first commit
This commit is contained in:
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
@@ -0,0 +1,337 @@
|
||||
import pickle
|
||||
|
||||
import numpy as np
|
||||
import numpy.testing as npt
|
||||
from numpy.testing import assert_allclose, assert_equal
|
||||
from pytest import raises as assert_raises
|
||||
|
||||
import numpy.ma.testutils as ma_npt
|
||||
|
||||
from scipy._lib._util import getfullargspec_no_self as _getfullargspec
|
||||
from scipy import stats
|
||||
|
||||
|
||||
def check_named_results(res, attributes, ma=False):
|
||||
for i, attr in enumerate(attributes):
|
||||
if ma:
|
||||
ma_npt.assert_equal(res[i], getattr(res, attr))
|
||||
else:
|
||||
npt.assert_equal(res[i], getattr(res, attr))
|
||||
|
||||
|
||||
def check_normalization(distfn, args, distname):
|
||||
norm_moment = distfn.moment(0, *args)
|
||||
npt.assert_allclose(norm_moment, 1.0)
|
||||
|
||||
# this is a temporary plug: either ncf or expect is problematic;
|
||||
# best be marked as a knownfail, but I've no clue how to do it.
|
||||
if distname == "ncf":
|
||||
atol, rtol = 1e-5, 0
|
||||
else:
|
||||
atol, rtol = 1e-7, 1e-7
|
||||
|
||||
normalization_expect = distfn.expect(lambda x: 1, args=args)
|
||||
npt.assert_allclose(normalization_expect, 1.0, atol=atol, rtol=rtol,
|
||||
err_msg=distname, verbose=True)
|
||||
|
||||
_a, _b = distfn.support(*args)
|
||||
normalization_cdf = distfn.cdf(_b, *args)
|
||||
npt.assert_allclose(normalization_cdf, 1.0)
|
||||
|
||||
|
||||
def check_moment(distfn, arg, m, v, msg):
|
||||
m1 = distfn.moment(1, *arg)
|
||||
m2 = distfn.moment(2, *arg)
|
||||
if not np.isinf(m):
|
||||
npt.assert_almost_equal(m1, m, decimal=10, err_msg=msg +
|
||||
' - 1st moment')
|
||||
else: # or np.isnan(m1),
|
||||
npt.assert_(np.isinf(m1),
|
||||
msg + ' - 1st moment -infinite, m1=%s' % str(m1))
|
||||
|
||||
if not np.isinf(v):
|
||||
npt.assert_almost_equal(m2 - m1 * m1, v, decimal=10, err_msg=msg +
|
||||
' - 2ndt moment')
|
||||
else: # or np.isnan(m2),
|
||||
npt.assert_(np.isinf(m2),
|
||||
msg + ' - 2nd moment -infinite, m2=%s' % str(m2))
|
||||
|
||||
|
||||
def check_mean_expect(distfn, arg, m, msg):
|
||||
if np.isfinite(m):
|
||||
m1 = distfn.expect(lambda x: x, arg)
|
||||
npt.assert_almost_equal(m1, m, decimal=5, err_msg=msg +
|
||||
' - 1st moment (expect)')
|
||||
|
||||
|
||||
def check_var_expect(distfn, arg, m, v, msg):
|
||||
if np.isfinite(v):
|
||||
m2 = distfn.expect(lambda x: x*x, arg)
|
||||
npt.assert_almost_equal(m2, v + m*m, decimal=5, err_msg=msg +
|
||||
' - 2st moment (expect)')
|
||||
|
||||
|
||||
def check_skew_expect(distfn, arg, m, v, s, msg):
|
||||
if np.isfinite(s):
|
||||
m3e = distfn.expect(lambda x: np.power(x-m, 3), arg)
|
||||
npt.assert_almost_equal(m3e, s * np.power(v, 1.5),
|
||||
decimal=5, err_msg=msg + ' - skew')
|
||||
else:
|
||||
npt.assert_(np.isnan(s))
|
||||
|
||||
|
||||
def check_kurt_expect(distfn, arg, m, v, k, msg):
|
||||
if np.isfinite(k):
|
||||
m4e = distfn.expect(lambda x: np.power(x-m, 4), arg)
|
||||
npt.assert_allclose(m4e, (k + 3.) * np.power(v, 2), atol=1e-5, rtol=1e-5,
|
||||
err_msg=msg + ' - kurtosis')
|
||||
elif not np.isposinf(k):
|
||||
npt.assert_(np.isnan(k))
|
||||
|
||||
|
||||
def check_entropy(distfn, arg, msg):
|
||||
ent = distfn.entropy(*arg)
|
||||
npt.assert_(not np.isnan(ent), msg + 'test Entropy is nan')
|
||||
|
||||
|
||||
def check_private_entropy(distfn, args, superclass):
|
||||
# compare a generic _entropy with the distribution-specific implementation
|
||||
npt.assert_allclose(distfn._entropy(*args),
|
||||
superclass._entropy(distfn, *args))
|
||||
|
||||
|
||||
def check_entropy_vect_scale(distfn, arg):
|
||||
# check 2-d
|
||||
sc = np.asarray([[1, 2], [3, 4]])
|
||||
v_ent = distfn.entropy(*arg, scale=sc)
|
||||
s_ent = [distfn.entropy(*arg, scale=s) for s in sc.ravel()]
|
||||
s_ent = np.asarray(s_ent).reshape(v_ent.shape)
|
||||
assert_allclose(v_ent, s_ent, atol=1e-14)
|
||||
|
||||
# check invalid value, check cast
|
||||
sc = [1, 2, -3]
|
||||
v_ent = distfn.entropy(*arg, scale=sc)
|
||||
s_ent = [distfn.entropy(*arg, scale=s) for s in sc]
|
||||
s_ent = np.asarray(s_ent).reshape(v_ent.shape)
|
||||
assert_allclose(v_ent, s_ent, atol=1e-14)
|
||||
|
||||
|
||||
def check_edge_support(distfn, args):
|
||||
# Make sure that x=self.a and self.b are handled correctly.
|
||||
x = distfn.support(*args)
|
||||
if isinstance(distfn, stats.rv_discrete):
|
||||
x = x[0]-1, x[1]
|
||||
|
||||
npt.assert_equal(distfn.cdf(x, *args), [0.0, 1.0])
|
||||
npt.assert_equal(distfn.sf(x, *args), [1.0, 0.0])
|
||||
|
||||
if distfn.name not in ('skellam', 'dlaplace'):
|
||||
# with a = -inf, log(0) generates warnings
|
||||
npt.assert_equal(distfn.logcdf(x, *args), [-np.inf, 0.0])
|
||||
npt.assert_equal(distfn.logsf(x, *args), [0.0, -np.inf])
|
||||
|
||||
npt.assert_equal(distfn.ppf([0.0, 1.0], *args), x)
|
||||
npt.assert_equal(distfn.isf([0.0, 1.0], *args), x[::-1])
|
||||
|
||||
# out-of-bounds for isf & ppf
|
||||
npt.assert_(np.isnan(distfn.isf([-1, 2], *args)).all())
|
||||
npt.assert_(np.isnan(distfn.ppf([-1, 2], *args)).all())
|
||||
|
||||
|
||||
def check_named_args(distfn, x, shape_args, defaults, meths):
|
||||
## Check calling w/ named arguments.
|
||||
|
||||
# check consistency of shapes, numargs and _parse signature
|
||||
signature = _getfullargspec(distfn._parse_args)
|
||||
npt.assert_(signature.varargs is None)
|
||||
npt.assert_(signature.varkw is None)
|
||||
npt.assert_(not signature.kwonlyargs)
|
||||
npt.assert_(list(signature.defaults) == list(defaults))
|
||||
|
||||
shape_argnames = signature.args[:-len(defaults)] # a, b, loc=0, scale=1
|
||||
if distfn.shapes:
|
||||
shapes_ = distfn.shapes.replace(',', ' ').split()
|
||||
else:
|
||||
shapes_ = ''
|
||||
npt.assert_(len(shapes_) == distfn.numargs)
|
||||
npt.assert_(len(shapes_) == len(shape_argnames))
|
||||
|
||||
# check calling w/ named arguments
|
||||
shape_args = list(shape_args)
|
||||
|
||||
vals = [meth(x, *shape_args) for meth in meths]
|
||||
npt.assert_(np.all(np.isfinite(vals)))
|
||||
|
||||
names, a, k = shape_argnames[:], shape_args[:], {}
|
||||
while names:
|
||||
k.update({names.pop(): a.pop()})
|
||||
v = [meth(x, *a, **k) for meth in meths]
|
||||
npt.assert_array_equal(vals, v)
|
||||
if 'n' not in k.keys():
|
||||
# `n` is first parameter of moment(), so can't be used as named arg
|
||||
npt.assert_equal(distfn.moment(1, *a, **k),
|
||||
distfn.moment(1, *shape_args))
|
||||
|
||||
# unknown arguments should not go through:
|
||||
k.update({'kaboom': 42})
|
||||
assert_raises(TypeError, distfn.cdf, x, **k)
|
||||
|
||||
|
||||
def check_random_state_property(distfn, args):
|
||||
# check the random_state attribute of a distribution *instance*
|
||||
|
||||
# This test fiddles with distfn.random_state. This breaks other tests,
|
||||
# hence need to save it and then restore.
|
||||
rndm = distfn.random_state
|
||||
|
||||
# baseline: this relies on the global state
|
||||
np.random.seed(1234)
|
||||
distfn.random_state = None
|
||||
r0 = distfn.rvs(*args, size=8)
|
||||
|
||||
# use an explicit instance-level random_state
|
||||
distfn.random_state = 1234
|
||||
r1 = distfn.rvs(*args, size=8)
|
||||
npt.assert_equal(r0, r1)
|
||||
|
||||
distfn.random_state = np.random.RandomState(1234)
|
||||
r2 = distfn.rvs(*args, size=8)
|
||||
npt.assert_equal(r0, r2)
|
||||
|
||||
# check that np.random.Generator can be used (numpy >= 1.17)
|
||||
if hasattr(np.random, 'default_rng'):
|
||||
# obtain a np.random.Generator object
|
||||
rng = np.random.default_rng(1234)
|
||||
distfn.rvs(*args, size=1, random_state=rng)
|
||||
|
||||
# can override the instance-level random_state for an individual .rvs call
|
||||
distfn.random_state = 2
|
||||
orig_state = distfn.random_state.get_state()
|
||||
|
||||
r3 = distfn.rvs(*args, size=8, random_state=np.random.RandomState(1234))
|
||||
npt.assert_equal(r0, r3)
|
||||
|
||||
# ... and that does not alter the instance-level random_state!
|
||||
npt.assert_equal(distfn.random_state.get_state(), orig_state)
|
||||
|
||||
# finally, restore the random_state
|
||||
distfn.random_state = rndm
|
||||
|
||||
|
||||
def check_meth_dtype(distfn, arg, meths):
|
||||
q0 = [0.25, 0.5, 0.75]
|
||||
x0 = distfn.ppf(q0, *arg)
|
||||
x_cast = [x0.astype(tp) for tp in
|
||||
(np.int_, np.float16, np.float32, np.float64)]
|
||||
|
||||
for x in x_cast:
|
||||
# casting may have clipped the values, exclude those
|
||||
distfn._argcheck(*arg)
|
||||
x = x[(distfn.a < x) & (x < distfn.b)]
|
||||
for meth in meths:
|
||||
val = meth(x, *arg)
|
||||
npt.assert_(val.dtype == np.float_)
|
||||
|
||||
|
||||
def check_ppf_dtype(distfn, arg):
|
||||
q0 = np.asarray([0.25, 0.5, 0.75])
|
||||
q_cast = [q0.astype(tp) for tp in (np.float16, np.float32, np.float64)]
|
||||
for q in q_cast:
|
||||
for meth in [distfn.ppf, distfn.isf]:
|
||||
val = meth(q, *arg)
|
||||
npt.assert_(val.dtype == np.float_)
|
||||
|
||||
|
||||
def check_cmplx_deriv(distfn, arg):
|
||||
# Distributions allow complex arguments.
|
||||
def deriv(f, x, *arg):
|
||||
x = np.asarray(x)
|
||||
h = 1e-10
|
||||
return (f(x + h*1j, *arg)/h).imag
|
||||
|
||||
x0 = distfn.ppf([0.25, 0.51, 0.75], *arg)
|
||||
x_cast = [x0.astype(tp) for tp in
|
||||
(np.int_, np.float16, np.float32, np.float64)]
|
||||
|
||||
for x in x_cast:
|
||||
# casting may have clipped the values, exclude those
|
||||
distfn._argcheck(*arg)
|
||||
x = x[(distfn.a < x) & (x < distfn.b)]
|
||||
|
||||
pdf, cdf, sf = distfn.pdf(x, *arg), distfn.cdf(x, *arg), distfn.sf(x, *arg)
|
||||
assert_allclose(deriv(distfn.cdf, x, *arg), pdf, rtol=1e-5)
|
||||
assert_allclose(deriv(distfn.logcdf, x, *arg), pdf/cdf, rtol=1e-5)
|
||||
|
||||
assert_allclose(deriv(distfn.sf, x, *arg), -pdf, rtol=1e-5)
|
||||
assert_allclose(deriv(distfn.logsf, x, *arg), -pdf/sf, rtol=1e-5)
|
||||
|
||||
assert_allclose(deriv(distfn.logpdf, x, *arg),
|
||||
deriv(distfn.pdf, x, *arg) / distfn.pdf(x, *arg),
|
||||
rtol=1e-5)
|
||||
|
||||
|
||||
def check_pickling(distfn, args):
|
||||
# check that a distribution instance pickles and unpickles
|
||||
# pay special attention to the random_state property
|
||||
|
||||
# save the random_state (restore later)
|
||||
rndm = distfn.random_state
|
||||
|
||||
# check unfrozen
|
||||
distfn.random_state = 1234
|
||||
distfn.rvs(*args, size=8)
|
||||
s = pickle.dumps(distfn)
|
||||
r0 = distfn.rvs(*args, size=8)
|
||||
|
||||
unpickled = pickle.loads(s)
|
||||
r1 = unpickled.rvs(*args, size=8)
|
||||
npt.assert_equal(r0, r1)
|
||||
|
||||
# also smoke test some methods
|
||||
medians = [distfn.ppf(0.5, *args), unpickled.ppf(0.5, *args)]
|
||||
npt.assert_equal(medians[0], medians[1])
|
||||
npt.assert_equal(distfn.cdf(medians[0], *args),
|
||||
unpickled.cdf(medians[1], *args))
|
||||
|
||||
# check frozen pickling/unpickling with rvs
|
||||
frozen_dist = distfn(*args)
|
||||
pkl = pickle.dumps(frozen_dist)
|
||||
unpickled = pickle.loads(pkl)
|
||||
|
||||
r0 = frozen_dist.rvs(size=8)
|
||||
r1 = unpickled.rvs(size=8)
|
||||
npt.assert_equal(r0, r1)
|
||||
|
||||
# check pickling/unpickling of .fit method
|
||||
if hasattr(distfn, "fit"):
|
||||
fit_function = distfn.fit
|
||||
pickled_fit_function = pickle.dumps(fit_function)
|
||||
unpickled_fit_function = pickle.loads(pickled_fit_function)
|
||||
assert fit_function.__name__ == unpickled_fit_function.__name__ == "fit"
|
||||
|
||||
# restore the random_state
|
||||
distfn.random_state = rndm
|
||||
|
||||
|
||||
def check_freezing(distfn, args):
|
||||
# regression test for gh-11089: freezing a distribution fails
|
||||
# if loc and/or scale are specified
|
||||
if isinstance(distfn, stats.rv_continuous):
|
||||
locscale = {'loc': 1, 'scale': 2}
|
||||
else:
|
||||
locscale = {'loc': 1}
|
||||
|
||||
rv = distfn(*args, **locscale)
|
||||
assert rv.a == distfn(*args).a
|
||||
assert rv.b == distfn(*args).b
|
||||
|
||||
|
||||
def check_rvs_broadcast(distfunc, distname, allargs, shape, shape_only, otype):
|
||||
np.random.seed(123)
|
||||
sample = distfunc.rvs(*allargs)
|
||||
assert_equal(sample.shape, shape, "%s: rvs failed to broadcast" % distname)
|
||||
if not shape_only:
|
||||
rvs = np.vectorize(lambda *allargs: distfunc.rvs(*allargs), otypes=otype)
|
||||
np.random.seed(123)
|
||||
expected = rvs(*allargs)
|
||||
assert_allclose(sample, expected, rtol=1e-13)
|
||||
@@ -0,0 +1,108 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: AtmWtAg (AtmWtAg.dat)
|
||||
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 41 to 47)
|
||||
Data (lines 61 to 108)
|
||||
|
||||
|
||||
Procedure: Analysis of Variance
|
||||
|
||||
|
||||
Reference: Powell, L.J., Murphy, T.J. and Gramlich, J.W. (1982).
|
||||
"The Absolute Isotopic Abundance & Atomic Weight
|
||||
of a Reference Sample of Silver".
|
||||
NBS Journal of Research, 87, pp. 9-19.
|
||||
|
||||
|
||||
Data: 1 Factor
|
||||
2 Treatments
|
||||
24 Replicates/Cell
|
||||
48 Observations
|
||||
7 Constant Leading Digits
|
||||
Average Level of Difficulty
|
||||
Observed Data
|
||||
|
||||
|
||||
Model: 3 Parameters (mu, tau_1, tau_2)
|
||||
y_{ij} = mu + tau_i + epsilon_{ij}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Certified Values:
|
||||
|
||||
Source of Sums of Mean
|
||||
Variation df Squares Squares F Statistic
|
||||
|
||||
|
||||
Between Instrument 1 3.63834187500000E-09 3.63834187500000E-09 1.59467335677930E+01
|
||||
Within Instrument 46 1.04951729166667E-08 2.28155932971014E-10
|
||||
|
||||
Certified R-Squared 2.57426544538321E-01
|
||||
|
||||
Certified Residual
|
||||
Standard Deviation 1.51048314446410E-05
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: Instrument AgWt
|
||||
1 107.8681568
|
||||
1 107.8681465
|
||||
1 107.8681572
|
||||
1 107.8681785
|
||||
1 107.8681446
|
||||
1 107.8681903
|
||||
1 107.8681526
|
||||
1 107.8681494
|
||||
1 107.8681616
|
||||
1 107.8681587
|
||||
1 107.8681519
|
||||
1 107.8681486
|
||||
1 107.8681419
|
||||
1 107.8681569
|
||||
1 107.8681508
|
||||
1 107.8681672
|
||||
1 107.8681385
|
||||
1 107.8681518
|
||||
1 107.8681662
|
||||
1 107.8681424
|
||||
1 107.8681360
|
||||
1 107.8681333
|
||||
1 107.8681610
|
||||
1 107.8681477
|
||||
2 107.8681079
|
||||
2 107.8681344
|
||||
2 107.8681513
|
||||
2 107.8681197
|
||||
2 107.8681604
|
||||
2 107.8681385
|
||||
2 107.8681642
|
||||
2 107.8681365
|
||||
2 107.8681151
|
||||
2 107.8681082
|
||||
2 107.8681517
|
||||
2 107.8681448
|
||||
2 107.8681198
|
||||
2 107.8681482
|
||||
2 107.8681334
|
||||
2 107.8681609
|
||||
2 107.8681101
|
||||
2 107.8681512
|
||||
2 107.8681469
|
||||
2 107.8681360
|
||||
2 107.8681254
|
||||
2 107.8681261
|
||||
2 107.8681450
|
||||
2 107.8681368
|
||||
@@ -0,0 +1,85 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: SiRstv (SiRstv.dat)
|
||||
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 41 to 47)
|
||||
Data (lines 61 to 85)
|
||||
|
||||
|
||||
Procedure: Analysis of Variance
|
||||
|
||||
|
||||
Reference: Ehrstein, James and Croarkin, M. Carroll.
|
||||
Unpublished NIST dataset.
|
||||
|
||||
|
||||
Data: 1 Factor
|
||||
5 Treatments
|
||||
5 Replicates/Cell
|
||||
25 Observations
|
||||
3 Constant Leading Digits
|
||||
Lower Level of Difficulty
|
||||
Observed Data
|
||||
|
||||
|
||||
Model: 6 Parameters (mu,tau_1, ... , tau_5)
|
||||
y_{ij} = mu + tau_i + epsilon_{ij}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Certified Values:
|
||||
|
||||
Source of Sums of Mean
|
||||
Variation df Squares Squares F Statistic
|
||||
|
||||
Between Instrument 4 5.11462616000000E-02 1.27865654000000E-02 1.18046237440255E+00
|
||||
Within Instrument 20 2.16636560000000E-01 1.08318280000000E-02
|
||||
|
||||
Certified R-Squared 1.90999039051129E-01
|
||||
|
||||
Certified Residual
|
||||
Standard Deviation 1.04076068334656E-01
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: Instrument Resistance
|
||||
1 196.3052
|
||||
1 196.1240
|
||||
1 196.1890
|
||||
1 196.2569
|
||||
1 196.3403
|
||||
2 196.3042
|
||||
2 196.3825
|
||||
2 196.1669
|
||||
2 196.3257
|
||||
2 196.0422
|
||||
3 196.1303
|
||||
3 196.2005
|
||||
3 196.2889
|
||||
3 196.0343
|
||||
3 196.1811
|
||||
4 196.2795
|
||||
4 196.1748
|
||||
4 196.1494
|
||||
4 196.1485
|
||||
4 195.9885
|
||||
5 196.2119
|
||||
5 196.1051
|
||||
5 196.1850
|
||||
5 196.0052
|
||||
5 196.2090
|
||||
@@ -0,0 +1,249 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: SmLs01 (SmLs01.dat)
|
||||
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 41 to 47)
|
||||
Data (lines 61 to 249)
|
||||
|
||||
|
||||
Procedure: Analysis of Variance
|
||||
|
||||
|
||||
Reference: Simon, Stephen D. and Lesage, James P. (1989).
|
||||
"Assessing the Accuracy of ANOVA Calculations in
|
||||
Statistical Software".
|
||||
Computational Statistics & Data Analysis, 8, pp. 325-332.
|
||||
|
||||
|
||||
Data: 1 Factor
|
||||
9 Treatments
|
||||
21 Replicates/Cell
|
||||
189 Observations
|
||||
1 Constant Leading Digit
|
||||
Lower Level of Difficulty
|
||||
Generated Data
|
||||
|
||||
|
||||
Model: 10 Parameters (mu,tau_1, ... , tau_9)
|
||||
y_{ij} = mu + tau_i + epsilon_{ij}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Certified Values:
|
||||
|
||||
Source of Sums of Mean
|
||||
Variation df Squares Squares F Statistic
|
||||
|
||||
Between Treatment 8 1.68000000000000E+00 2.10000000000000E-01 2.10000000000000E+01
|
||||
Within Treatment 180 1.80000000000000E+00 1.00000000000000E-02
|
||||
|
||||
Certified R-Squared 4.82758620689655E-01
|
||||
|
||||
Certified Residual
|
||||
Standard Deviation 1.00000000000000E-01
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: Treatment Response
|
||||
1 1.4
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
1 1.3
|
||||
1 1.5
|
||||
2 1.3
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
2 1.2
|
||||
2 1.4
|
||||
3 1.5
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
3 1.4
|
||||
3 1.6
|
||||
4 1.3
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
4 1.2
|
||||
4 1.4
|
||||
5 1.5
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
5 1.4
|
||||
5 1.6
|
||||
6 1.3
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
6 1.2
|
||||
6 1.4
|
||||
7 1.5
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
7 1.4
|
||||
7 1.6
|
||||
8 1.3
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
8 1.2
|
||||
8 1.4
|
||||
9 1.5
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
9 1.4
|
||||
9 1.6
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,249 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: SmLs04 (SmLs04.dat)
|
||||
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 41 to 47)
|
||||
Data (lines 61 to 249)
|
||||
|
||||
|
||||
Procedure: Analysis of Variance
|
||||
|
||||
|
||||
Reference: Simon, Stephen D. and Lesage, James P. (1989).
|
||||
"Assessing the Accuracy of ANOVA Calculations in
|
||||
Statistical Software".
|
||||
Computational Statistics & Data Analysis, 8, pp. 325-332.
|
||||
|
||||
|
||||
Data: 1 Factor
|
||||
9 Treatments
|
||||
21 Replicates/Cell
|
||||
189 Observations
|
||||
7 Constant Leading Digits
|
||||
Average Level of Difficulty
|
||||
Generated Data
|
||||
|
||||
|
||||
Model: 10 Parameters (mu,tau_1, ... , tau_9)
|
||||
y_{ij} = mu + tau_i + epsilon_{ij}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Certified Values:
|
||||
|
||||
Source of Sums of Mean
|
||||
Variation df Squares Squares F Statistic
|
||||
|
||||
Between Treatment 8 1.68000000000000E+00 2.10000000000000E-01 2.10000000000000E+01
|
||||
Within Treatment 180 1.80000000000000E+00 1.00000000000000E-02
|
||||
|
||||
Certified R-Squared 4.82758620689655E-01
|
||||
|
||||
Certified Residual
|
||||
Standard Deviation 1.00000000000000E-01
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: Treatment Response
|
||||
1 1000000.4
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
1 1000000.3
|
||||
1 1000000.5
|
||||
2 1000000.3
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
2 1000000.2
|
||||
2 1000000.4
|
||||
3 1000000.5
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
3 1000000.4
|
||||
3 1000000.6
|
||||
4 1000000.3
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
4 1000000.2
|
||||
4 1000000.4
|
||||
5 1000000.5
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
5 1000000.4
|
||||
5 1000000.6
|
||||
6 1000000.3
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
6 1000000.2
|
||||
6 1000000.4
|
||||
7 1000000.5
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
7 1000000.4
|
||||
7 1000000.6
|
||||
8 1000000.3
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
8 1000000.2
|
||||
8 1000000.4
|
||||
9 1000000.5
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
9 1000000.4
|
||||
9 1000000.6
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,249 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: SmLs07 (SmLs07.dat)
|
||||
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 41 to 47)
|
||||
Data (lines 61 to 249)
|
||||
|
||||
|
||||
Procedure: Analysis of Variance
|
||||
|
||||
|
||||
Reference: Simon, Stephen D. and Lesage, James P. (1989).
|
||||
"Assessing the Accuracy of ANOVA Calculations in
|
||||
Statistical Software".
|
||||
Computational Statistics & Data Analysis, 8, pp. 325-332.
|
||||
|
||||
|
||||
Data: 1 Factor
|
||||
9 Treatments
|
||||
21 Replicates/Cell
|
||||
189 Observations
|
||||
13 Constant Leading Digits
|
||||
Higher Level of Difficulty
|
||||
Generated Data
|
||||
|
||||
|
||||
Model: 10 Parameters (mu,tau_1, ... , tau_9)
|
||||
y_{ij} = mu + tau_i + epsilon_{ij}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Certified Values:
|
||||
|
||||
Source of Sums of Mean
|
||||
Variation df Squares Squares F Statistic
|
||||
|
||||
Between Treatment 8 1.68000000000000E+00 2.10000000000000E-01 2.10000000000000E+01
|
||||
Within Treatment 180 1.80000000000000E+00 1.00000000000000E-02
|
||||
|
||||
Certified R-Squared 4.82758620689655E-01
|
||||
|
||||
Certified Residual
|
||||
Standard Deviation 1.00000000000000E-01
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: Treatment Response
|
||||
1 1000000000000.4
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
1 1000000000000.3
|
||||
1 1000000000000.5
|
||||
2 1000000000000.3
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
2 1000000000000.2
|
||||
2 1000000000000.4
|
||||
3 1000000000000.5
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
3 1000000000000.4
|
||||
3 1000000000000.6
|
||||
4 1000000000000.3
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
4 1000000000000.2
|
||||
4 1000000000000.4
|
||||
5 1000000000000.5
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
5 1000000000000.4
|
||||
5 1000000000000.6
|
||||
6 1000000000000.3
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
6 1000000000000.2
|
||||
6 1000000000000.4
|
||||
7 1000000000000.5
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
7 1000000000000.4
|
||||
7 1000000000000.6
|
||||
8 1000000000000.3
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
8 1000000000000.2
|
||||
8 1000000000000.4
|
||||
9 1000000000000.5
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
9 1000000000000.4
|
||||
9 1000000000000.6
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,97 @@
|
||||
NIST/ITL StRD
|
||||
Dataset Name: Norris (Norris.dat)
|
||||
|
||||
File Format: ASCII
|
||||
Certified Values (lines 31 to 46)
|
||||
Data (lines 61 to 96)
|
||||
|
||||
Procedure: Linear Least Squares Regression
|
||||
|
||||
Reference: Norris, J., NIST.
|
||||
Calibration of Ozone Monitors.
|
||||
|
||||
Data: 1 Response Variable (y)
|
||||
1 Predictor Variable (x)
|
||||
36 Observations
|
||||
Lower Level of Difficulty
|
||||
Observed Data
|
||||
|
||||
Model: Linear Class
|
||||
2 Parameters (B0,B1)
|
||||
|
||||
y = B0 + B1*x + e
|
||||
|
||||
|
||||
|
||||
Certified Regression Statistics
|
||||
|
||||
Standard Deviation
|
||||
Parameter Estimate of Estimate
|
||||
|
||||
B0 -0.262323073774029 0.232818234301152
|
||||
B1 1.00211681802045 0.429796848199937E-03
|
||||
|
||||
Residual
|
||||
Standard Deviation 0.884796396144373
|
||||
|
||||
R-Squared 0.999993745883712
|
||||
|
||||
|
||||
Certified Analysis of Variance Table
|
||||
|
||||
Source of Degrees of Sums of Mean
|
||||
Variation Freedom Squares Squares F Statistic
|
||||
|
||||
Regression 1 4255954.13232369 4255954.13232369 5436385.54079785
|
||||
Residual 34 26.6173985294224 0.782864662630069
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Data: y x
|
||||
0.1 0.2
|
||||
338.8 337.4
|
||||
118.1 118.2
|
||||
888.0 884.6
|
||||
9.2 10.1
|
||||
228.1 226.5
|
||||
668.5 666.3
|
||||
998.5 996.3
|
||||
449.1 448.6
|
||||
778.9 777.0
|
||||
559.2 558.2
|
||||
0.3 0.4
|
||||
0.1 0.6
|
||||
778.1 775.5
|
||||
668.8 666.9
|
||||
339.3 338.0
|
||||
448.9 447.5
|
||||
10.8 11.6
|
||||
557.7 556.0
|
||||
228.3 228.1
|
||||
998.0 995.8
|
||||
888.8 887.6
|
||||
119.6 120.2
|
||||
0.3 0.3
|
||||
0.6 0.3
|
||||
557.6 556.8
|
||||
339.3 339.1
|
||||
888.0 887.2
|
||||
998.5 999.0
|
||||
778.9 779.0
|
||||
10.2 11.1
|
||||
117.6 118.3
|
||||
228.9 229.2
|
||||
668.4 669.1
|
||||
449.2 448.9
|
||||
0.2 0.5
|
||||
|
||||
Binary file not shown.
Binary file not shown.
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,252 @@
|
||||
# To run this script, run
|
||||
# `python studentized_range_mpmath_ref.py`
|
||||
# in the "scipy/stats/tests/" directory
|
||||
|
||||
# This script generates a JSON file "./data/studentized_range_mpmath_ref.json"
|
||||
# that is used to compare the accuracy of `studentized_range` functions against
|
||||
# precise (20 DOP) results generated using `mpmath`.
|
||||
|
||||
# Equations in this file have been taken from
|
||||
# https://en.wikipedia.org/wiki/Studentized_range_distribution
|
||||
# and have been checked against the following reference:
|
||||
# Lund, R. E., and J. R. Lund. "Algorithm AS 190: Probabilities and
|
||||
# Upper Quantiles for the Studentized Range." Journal of the Royal
|
||||
# Statistical Society. Series C (Applied Statistics), vol. 32, no. 2,
|
||||
# 1983, pp. 204-210. JSTOR, www.jstor.org/stable/2347300. Accessed 18
|
||||
# Feb. 2021.
|
||||
|
||||
# Note: I would have prefered to use pickle rather than JSON, but -
|
||||
# due to security concerns - decided against it.
|
||||
import itertools
|
||||
from collections import namedtuple
|
||||
import json
|
||||
import time
|
||||
|
||||
import os
|
||||
from multiprocessing import Pool, cpu_count
|
||||
|
||||
from mpmath import gamma, pi, sqrt, quad, inf, mpf, mp
|
||||
from mpmath import npdf as phi
|
||||
from mpmath import ncdf as Phi
|
||||
|
||||
results_filepath = "data/studentized_range_mpmath_ref.json"
|
||||
num_pools = max(cpu_count() - 1, 1)
|
||||
|
||||
MPResult = namedtuple("MPResult", ["src_case", "mp_result"])
|
||||
|
||||
CdfCase = namedtuple("CdfCase",
|
||||
["q", "k", "v", "expected_atol", "expected_rtol"])
|
||||
|
||||
MomentCase = namedtuple("MomentCase",
|
||||
["m", "k", "v", "expected_atol", "expected_rtol"])
|
||||
|
||||
# Load previously generated JSON results, or init a new dict if none exist
|
||||
if os.path.isfile(results_filepath):
|
||||
res_dict = json.load(open(results_filepath, mode="r"))
|
||||
else:
|
||||
res_dict = dict()
|
||||
|
||||
# Frame out data structure. Store data with the function type as a top level
|
||||
# key to allow future expansion
|
||||
res_dict["COMMENT"] = ("!!!!!! THIS FILE WAS AUTOGENERATED BY RUNNING "
|
||||
"`python studentized_range_mpmath_ref.py` !!!!!!")
|
||||
res_dict.setdefault("cdf_data", [])
|
||||
res_dict.setdefault("pdf_data", [])
|
||||
res_dict.setdefault("moment_data", [])
|
||||
|
||||
general_atol, general_rtol = 1e-11, 1e-11
|
||||
|
||||
mp.dps = 24
|
||||
|
||||
cp_q = [0.1, 1, 4, 10]
|
||||
cp_k = [3, 10, 20]
|
||||
cp_nu = [3, 10, 20, 50, 100, 120]
|
||||
|
||||
cdf_pdf_cases = [
|
||||
CdfCase(*case,
|
||||
general_atol,
|
||||
general_rtol)
|
||||
for case in
|
||||
itertools.product(cp_q, cp_k, cp_nu)
|
||||
]
|
||||
|
||||
mom_atol, mom_rtol = 1e-9, 1e-9
|
||||
# These are EXTREMELY slow - Multiple days each in worst case.
|
||||
moment_cases = [
|
||||
MomentCase(i, 3, 10, mom_atol, mom_rtol)
|
||||
for i in range(5)
|
||||
]
|
||||
|
||||
|
||||
def write_data():
|
||||
"""Writes the current res_dict to the target JSON file"""
|
||||
with open(results_filepath, mode="w") as f:
|
||||
json.dump(res_dict, f, indent=2)
|
||||
|
||||
|
||||
def to_dict(named_tuple):
|
||||
"""Converts a namedtuple to a dict"""
|
||||
return dict(named_tuple._asdict())
|
||||
|
||||
|
||||
def mp_res_to_dict(mp_result):
|
||||
"""Formats an MPResult namedtuple into a dict for JSON dumping"""
|
||||
return {
|
||||
"src_case": to_dict(mp_result.src_case),
|
||||
|
||||
# np assert can't handle mpf, so take the accuracy hit here.
|
||||
"mp_result": float(mp_result.mp_result)
|
||||
}
|
||||
|
||||
|
||||
def cdf_mp(q, k, nu):
|
||||
"""Straightforward implementation of studentized range CDF"""
|
||||
q, k, nu = mpf(q), mpf(k), mpf(nu)
|
||||
|
||||
def inner(s, z):
|
||||
return phi(z) * (Phi(z + q * s) - Phi(z)) ** (k - 1)
|
||||
|
||||
def outer(s, z):
|
||||
return s ** (nu - 1) * phi(sqrt(nu) * s) * inner(s, z)
|
||||
|
||||
def whole(s, z):
|
||||
return (sqrt(2 * pi) * k * nu ** (nu / 2)
|
||||
/ (gamma(nu / 2) * 2 ** (nu / 2 - 1)) * outer(s, z))
|
||||
|
||||
res = quad(whole, [0, inf], [-inf, inf],
|
||||
method="gauss-legendre", maxdegree=10)
|
||||
return res
|
||||
|
||||
|
||||
def pdf_mp(q, k, nu):
|
||||
"""Straightforward implementation of studentized range PDF"""
|
||||
q, k, nu = mpf(q), mpf(k), mpf(nu)
|
||||
|
||||
def inner(s, z):
|
||||
return phi(z + q * s) * phi(z) * (Phi(z + q * s) - Phi(z)) ** (k - 2)
|
||||
|
||||
def outer(s, z):
|
||||
return s ** nu * phi(sqrt(nu) * s) * inner(s, z)
|
||||
|
||||
def whole(s, z):
|
||||
return (sqrt(2 * pi) * k * (k - 1) * nu ** (nu / 2)
|
||||
/ (gamma(nu / 2) * 2 ** (nu / 2 - 1)) * outer(s, z))
|
||||
|
||||
res = quad(whole, [0, inf], [-inf, inf],
|
||||
method="gauss-legendre", maxdegree=10)
|
||||
return res
|
||||
|
||||
|
||||
def moment_mp(m, k, nu):
|
||||
"""Implementation of the studentized range moment"""
|
||||
m, k, nu = mpf(m), mpf(k), mpf(nu)
|
||||
|
||||
def inner(q, s, z):
|
||||
return phi(z + q * s) * phi(z) * (Phi(z + q * s) - Phi(z)) ** (k - 2)
|
||||
|
||||
def outer(q, s, z):
|
||||
return s ** nu * phi(sqrt(nu) * s) * inner(q, s, z)
|
||||
|
||||
def pdf(q, s, z):
|
||||
return (sqrt(2 * pi) * k * (k - 1) * nu ** (nu / 2)
|
||||
/ (gamma(nu / 2) * 2 ** (nu / 2 - 1)) * outer(q, s, z))
|
||||
|
||||
def whole(q, s, z):
|
||||
return q ** m * pdf(q, s, z)
|
||||
|
||||
res = quad(whole, [0, inf], [0, inf], [-inf, inf],
|
||||
method="gauss-legendre", maxdegree=10)
|
||||
return res
|
||||
|
||||
|
||||
def result_exists(set_key, case):
|
||||
"""Searches the results dict for a result in the set that matches a case.
|
||||
Returns True if such a case exists."""
|
||||
if set_key not in res_dict:
|
||||
raise ValueError(f"{set_key} not present in data structure!")
|
||||
|
||||
case_dict = to_dict(case)
|
||||
existing_res = list(filter(
|
||||
lambda res: res["src_case"] == case_dict, # dict comparison
|
||||
res_dict[set_key]))
|
||||
|
||||
return len(existing_res) > 0
|
||||
|
||||
|
||||
def run(case, run_lambda, set_key, index=0, total_cases=0):
|
||||
"""Runs the single passed case, returning an mp dictionary and index"""
|
||||
t_start = time.perf_counter()
|
||||
|
||||
res = run_lambda(case)
|
||||
|
||||
print(f"Finished {index + 1}/{total_cases} in batch. "
|
||||
f"(Took {time.perf_counter() - t_start}s)")
|
||||
|
||||
return index, set_key, mp_res_to_dict(MPResult(case, res))
|
||||
|
||||
|
||||
def write_result(res):
|
||||
"""A callback for completed jobs. Inserts and writes a calculated result
|
||||
to file."""
|
||||
index, set_key, result_dict = res
|
||||
res_dict[set_key].insert(index, result_dict)
|
||||
write_data()
|
||||
|
||||
|
||||
def run_cases(cases, run_lambda, set_key):
|
||||
"""Runs an array of cases and writes to file"""
|
||||
# Generate jobs to run from cases that do not have a result in
|
||||
# the previously loaded JSON.
|
||||
job_arg = [(case, run_lambda, set_key, index, len(cases))
|
||||
for index, case in enumerate(cases)
|
||||
if not result_exists(set_key, case)]
|
||||
|
||||
print(f"{len(cases) - len(job_arg)}/{len(cases)} cases won't be "
|
||||
f"calculated because their results already exist.")
|
||||
|
||||
jobs = []
|
||||
pool = Pool(num_pools)
|
||||
|
||||
# Run all using multiprocess
|
||||
for case in job_arg:
|
||||
jobs.append(pool.apply_async(run, args=case, callback=write_result))
|
||||
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
|
||||
def run_pdf(case):
|
||||
return pdf_mp(case.q, case.k, case.v)
|
||||
|
||||
|
||||
def run_cdf(case):
|
||||
return cdf_mp(case.q, case.k, case.v)
|
||||
|
||||
|
||||
def run_moment(case):
|
||||
return moment_mp(case.m, case.k, case.v)
|
||||
|
||||
|
||||
def main():
|
||||
t_start = time.perf_counter()
|
||||
|
||||
total_cases = 2 * len(cdf_pdf_cases) + len(moment_cases)
|
||||
print(f"Processing {total_cases} test cases")
|
||||
|
||||
print(f"Running 1st batch ({len(cdf_pdf_cases)} PDF cases). "
|
||||
f"These take about 30s each.")
|
||||
run_cases(cdf_pdf_cases, run_pdf, "pdf_data")
|
||||
|
||||
print(f"Running 2nd batch ({len(cdf_pdf_cases)} CDF cases). "
|
||||
f"These take about 30s each.")
|
||||
run_cases(cdf_pdf_cases, run_cdf, "cdf_data")
|
||||
|
||||
print(f"Running 3rd batch ({len(moment_cases)} moment cases). "
|
||||
f"These take about anywhere from a few hours to days each.")
|
||||
run_cases(moment_cases, run_moment, "moment_data")
|
||||
|
||||
print(f"Test data generated in {time.perf_counter() - t_start}s")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,541 @@
|
||||
# Many scipy.stats functions support `axis` and `nan_policy` parameters.
|
||||
# When the two are combined, it can be tricky to get all the behavior just
|
||||
# right. This file contains a suite of common tests for scipy.stats functions
|
||||
# that support `axis` and `nan_policy` and additional tests for some associated
|
||||
# functions in stats._util.
|
||||
|
||||
from itertools import product, combinations_with_replacement
|
||||
import re
|
||||
import pickle
|
||||
import pytest
|
||||
|
||||
import numpy as np
|
||||
from numpy.lib import NumpyVersion
|
||||
from numpy.testing import assert_allclose, assert_equal
|
||||
from scipy import stats
|
||||
|
||||
|
||||
axis_nan_policy_cases = [
|
||||
# function, args, kwds, number of samples, paired, unpacker function
|
||||
# args, kwds typically aren't needed; just showing that they work
|
||||
(stats.kruskal, tuple(), dict(), 3, False, None), # 4 samples is slow
|
||||
(stats.ranksums, ('less',), dict(), 2, False, None),
|
||||
(stats.mannwhitneyu, tuple(), {'method': 'asymptotic'}, 2, False, None),
|
||||
(stats.wilcoxon, ('pratt',), {'mode': 'auto'}, 2, True, None),
|
||||
(stats.wilcoxon, tuple(), dict(), 1, True, None),
|
||||
]
|
||||
|
||||
# If the message is one of those expected, put nans in
|
||||
# appropriate places of `statistics` and `pvalues`
|
||||
too_small_messages = {"The input contains nan", # for nan_policy="raise"
|
||||
"Degrees of freedom <= 0 for slice",
|
||||
"x and y should have at least 5 elements",
|
||||
"Data must be at least length 3",
|
||||
"The sample must contain at least two",
|
||||
"x and y must contain at least two",
|
||||
"division by zero",
|
||||
"Mean of empty slice",
|
||||
"Data passed to ks_2samp must not be empty",
|
||||
"Not enough test observations",
|
||||
"Not enough other observations",
|
||||
"At least one observation is required",
|
||||
"zero-size array to reduction operation maximum",
|
||||
"`x` and `y` must be of nonzero size.",
|
||||
"The exact distribution of the Wilcoxon test"}
|
||||
|
||||
|
||||
def _mixed_data_generator(n_samples, n_repetitions, axis, rng,
|
||||
paired=False):
|
||||
# generate random samples to check the response of hypothesis tests to
|
||||
# samples with different (but broadcastable) shapes and various
|
||||
# nan patterns (e.g. all nans, some nans, no nans) along axis-slices
|
||||
|
||||
data = []
|
||||
for i in range(n_samples):
|
||||
n_patterns = 6 # number of distinct nan patterns
|
||||
n_obs = 20 if paired else 20 + i # observations per axis-slice
|
||||
x = np.ones((n_repetitions, n_patterns, n_obs)) * np.nan
|
||||
|
||||
for j in range(n_repetitions):
|
||||
samples = x[j, :, :]
|
||||
|
||||
# case 0: axis-slice with all nans (0 reals)
|
||||
# cases 1-3: axis-slice with 1-3 reals (the rest nans)
|
||||
# case 4: axis-slice with mostly (all but two) reals
|
||||
# case 5: axis slice with all reals
|
||||
for k, n_reals in enumerate([0, 1, 2, 3, n_obs-2, n_obs]):
|
||||
# for cases 1-3, need paired nansw to be in the same place
|
||||
indices = rng.permutation(n_obs)[:n_reals]
|
||||
samples[k, indices] = rng.random(size=n_reals)
|
||||
|
||||
# permute the axis-slices just to show that order doesn't matter
|
||||
samples[:] = rng.permutation(samples, axis=0)
|
||||
|
||||
# For multi-sample tests, we want to test broadcasting and check
|
||||
# that nan policy works correctly for each nan pattern for each input.
|
||||
# This takes care of both simultaneosly.
|
||||
new_shape = [n_repetitions] + [1]*n_samples + [n_obs]
|
||||
new_shape[1 + i] = 6
|
||||
x = x.reshape(new_shape)
|
||||
|
||||
x = np.moveaxis(x, -1, axis)
|
||||
data.append(x)
|
||||
return data
|
||||
|
||||
|
||||
def _homogeneous_data_generator(n_samples, n_repetitions, axis, rng,
|
||||
paired=False, all_nans=True):
|
||||
# generate random samples to check the response of hypothesis tests to
|
||||
# samples with different (but broadcastable) shapes and homogeneous
|
||||
# data (all nans or all finite)
|
||||
data = []
|
||||
for i in range(n_samples):
|
||||
n_obs = 20 if paired else 20 + i # observations per axis-slice
|
||||
shape = [n_repetitions] + [1]*n_samples + [n_obs]
|
||||
shape[1 + i] = 2
|
||||
x = np.ones(shape) * np.nan if all_nans else rng.random(shape)
|
||||
x = np.moveaxis(x, -1, axis)
|
||||
data.append(x)
|
||||
return data
|
||||
|
||||
|
||||
def nan_policy_1d(hypotest, data1d, unpacker, *args,
|
||||
nan_policy='raise', paired=False, _no_deco=True, **kwds):
|
||||
# Reference implementation for how `nan_policy` should work for 1d samples
|
||||
|
||||
if nan_policy == 'raise':
|
||||
for sample in data1d:
|
||||
if np.any(np.isnan(sample)):
|
||||
raise ValueError("The input contains nan values")
|
||||
|
||||
elif nan_policy == 'propagate':
|
||||
# For all hypothesis tests tested, returning nans is the right thing.
|
||||
# But many hypothesis tests don't propagate correctly (e.g. they treat
|
||||
# np.nan the same as np.inf, which doesn't make sense when ranks are
|
||||
# involved) so override that behavior here.
|
||||
for sample in data1d:
|
||||
if np.any(np.isnan(sample)):
|
||||
return np.nan, np.nan
|
||||
|
||||
elif nan_policy == 'omit':
|
||||
# manually omit nans (or pairs in which at least one element is nan)
|
||||
if not paired:
|
||||
data1d = [sample[~np.isnan(sample)] for sample in data1d]
|
||||
else:
|
||||
nan_mask = np.isnan(data1d[0])
|
||||
for sample in data1d[1:]:
|
||||
nan_mask = np.logical_or(nan_mask, np.isnan(sample))
|
||||
data1d = [sample[~nan_mask] for sample in data1d]
|
||||
|
||||
return unpacker(hypotest(*data1d, *args, _no_deco=_no_deco, **kwds))
|
||||
|
||||
|
||||
@pytest.mark.parametrize(("hypotest", "args", "kwds", "n_samples", "paired",
|
||||
"unpacker"), axis_nan_policy_cases)
|
||||
@pytest.mark.parametrize(("nan_policy"), ("propagate", "omit", "raise"))
|
||||
@pytest.mark.parametrize(("axis"), (1,))
|
||||
@pytest.mark.parametrize(("data_generator"), ("mixed",))
|
||||
def test_axis_nan_policy_fast(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, axis,
|
||||
data_generator):
|
||||
_axis_nan_policy_test(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, axis, data_generator)
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
@pytest.mark.parametrize(("hypotest", "args", "kwds", "n_samples", "paired",
|
||||
"unpacker"), axis_nan_policy_cases)
|
||||
@pytest.mark.parametrize(("nan_policy"), ("propagate", "omit", "raise"))
|
||||
@pytest.mark.parametrize(("axis"), range(-3, 3))
|
||||
@pytest.mark.parametrize(("data_generator"),
|
||||
("all_nans", "all_finite", "mixed"))
|
||||
def test_axis_nan_policy_full(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, axis,
|
||||
data_generator):
|
||||
_axis_nan_policy_test(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, axis, data_generator)
|
||||
|
||||
|
||||
def _axis_nan_policy_test(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, axis, data_generator):
|
||||
# Tests the 1D and vectorized behavior of hypothesis tests against a
|
||||
# reference implementation (nan_policy_1d with np.ndenumerate)
|
||||
|
||||
# Some hypothesis tests return a non-iterable that needs an `unpacker` to
|
||||
# extract the statistic and p-value. For those that don't:
|
||||
if not unpacker:
|
||||
def unpacker(res):
|
||||
return res
|
||||
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
pytest.xfail("Generator `permutation` method doesn't support `axis`")
|
||||
rng = np.random.default_rng(0)
|
||||
|
||||
# Generate multi-dimensional test data with all important combinations
|
||||
# of patterns of nans along `axis`
|
||||
n_repetitions = 3 # number of repetitions of each pattern
|
||||
data_gen_kwds = {'n_samples': n_samples, 'n_repetitions': n_repetitions,
|
||||
'axis': axis, 'rng': rng, 'paired': paired}
|
||||
if data_generator == 'mixed':
|
||||
inherent_size = 6 # number of distinct types of patterns
|
||||
data = _mixed_data_generator(**data_gen_kwds)
|
||||
elif data_generator == 'all_nans':
|
||||
inherent_size = 2 # hard-coded in _homogeneous_data_generator
|
||||
data_gen_kwds['all_nans'] = True
|
||||
data = _homogeneous_data_generator(**data_gen_kwds)
|
||||
elif data_generator == 'all_finite':
|
||||
inherent_size = 2 # hard-coded in _homogeneous_data_generator
|
||||
data_gen_kwds['all_nans'] = False
|
||||
data = _homogeneous_data_generator(**data_gen_kwds)
|
||||
|
||||
output_shape = [n_repetitions] + [inherent_size]*n_samples
|
||||
|
||||
# To generate reference behavior to compare against, loop over the axis-
|
||||
# slices in data. Make indexing easier by moving `axis` to the end and
|
||||
# broadcasting all samples to the same shape.
|
||||
data_b = [np.moveaxis(sample, axis, -1) for sample in data]
|
||||
data_b = [np.broadcast_to(sample, output_shape + [sample.shape[-1]])
|
||||
for sample in data_b]
|
||||
statistics = np.zeros(output_shape)
|
||||
pvalues = np.zeros(output_shape)
|
||||
|
||||
for i, _ in np.ndenumerate(statistics):
|
||||
data1d = [sample[i] for sample in data_b]
|
||||
with np.errstate(divide='ignore', invalid='ignore'):
|
||||
try:
|
||||
res1d = nan_policy_1d(hypotest, data1d, unpacker, *args,
|
||||
nan_policy=nan_policy, paired=paired,
|
||||
_no_deco=True, **kwds)
|
||||
|
||||
# Eventually we'll check the results of a single, vectorized
|
||||
# call of `hypotest` against the arrays `statistics` and
|
||||
# `pvalues` populated using the reference `nan_policy_1d`.
|
||||
# But while we're at it, check the results of a 1D call to
|
||||
# `hypotest` against the reference `nan_policy_1d`.
|
||||
res1db = unpacker(hypotest(*data1d, *args,
|
||||
nan_policy=nan_policy, **kwds))
|
||||
assert_equal(res1db[0], res1d[0])
|
||||
if len(res1db) == 2:
|
||||
assert_equal(res1db[1], res1d[1])
|
||||
|
||||
# When there is not enough data in 1D samples, many existing
|
||||
# hypothesis tests raise errors instead of returning nans .
|
||||
# For vectorized calls, we put nans in the corresponding elements
|
||||
# of the output.
|
||||
except (RuntimeWarning, ValueError, ZeroDivisionError) as e:
|
||||
|
||||
# whatever it is, make sure same error is raised by both
|
||||
# `nan_policy_1d` and `hypotest`
|
||||
with pytest.raises(type(e), match=re.escape(str(e))):
|
||||
nan_policy_1d(hypotest, data1d, unpacker, *args,
|
||||
nan_policy=nan_policy, paired=paired,
|
||||
_no_deco=True, **kwds)
|
||||
with pytest.raises(type(e), match=re.escape(str(e))):
|
||||
hypotest(*data1d, *args, nan_policy=nan_policy, **kwds)
|
||||
|
||||
if any([str(e).startswith(message)
|
||||
for message in too_small_messages]):
|
||||
res1d = np.nan, np.nan
|
||||
else:
|
||||
raise e
|
||||
statistics[i] = res1d[0]
|
||||
if len(res1d) == 2:
|
||||
pvalues[i] = res1d[1]
|
||||
|
||||
# Perform a vectorized call to the hypothesis test.
|
||||
# If `nan_policy == 'raise'`, check that it raises the appropriate error.
|
||||
# If not, compare against the output against `statistics` and `pvalues`
|
||||
if nan_policy == 'raise' and not data_generator == "all_finite":
|
||||
message = 'The input contains nan values'
|
||||
with pytest.raises(ValueError, match=message):
|
||||
hypotest(*data, axis=axis, nan_policy=nan_policy, *args, **kwds)
|
||||
|
||||
else:
|
||||
with np.errstate(divide='ignore', invalid='ignore'):
|
||||
res = unpacker(hypotest(*data, axis=axis, nan_policy=nan_policy,
|
||||
*args, **kwds))
|
||||
|
||||
assert_equal(res[0], statistics)
|
||||
assert_equal(res[0].dtype, statistics.dtype)
|
||||
if len(res) == 2:
|
||||
assert_equal(res[1], pvalues)
|
||||
assert_equal(res[1].dtype, pvalues.dtype)
|
||||
|
||||
|
||||
@pytest.mark.parametrize(("hypotest", "args", "kwds", "n_samples", "paired",
|
||||
"unpacker"), axis_nan_policy_cases)
|
||||
@pytest.mark.parametrize(("nan_policy"), ("propagate", "omit", "raise"))
|
||||
@pytest.mark.parametrize(("data_generator"),
|
||||
("all_nans", "all_finite", "mixed", "empty"))
|
||||
def test_axis_nan_policy_axis_is_None(hypotest, args, kwds, n_samples, paired,
|
||||
unpacker, nan_policy, data_generator):
|
||||
# check for correct behavior when `axis=None`
|
||||
|
||||
if not unpacker:
|
||||
def unpacker(res):
|
||||
return res
|
||||
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
pytest.xfail("Generator `permutation` method doesn't support `axis`")
|
||||
rng = np.random.default_rng(0)
|
||||
|
||||
if data_generator == "empty":
|
||||
data = [rng.random((2, 0)) for i in range(n_samples)]
|
||||
else:
|
||||
data = [rng.random((2, 20)) for i in range(n_samples)]
|
||||
|
||||
if data_generator == "mixed":
|
||||
masks = [rng.random((2, 20)) > 0.9 for i in range(n_samples)]
|
||||
for sample, mask in zip(data, masks):
|
||||
sample[mask] = np.nan
|
||||
elif data_generator == "all_nans":
|
||||
data = [sample * np.nan for sample in data]
|
||||
|
||||
data_raveled = [sample.ravel() for sample in data]
|
||||
|
||||
if nan_policy == 'raise' and data_generator not in {"all_finite", "empty"}:
|
||||
message = 'The input contains nan values'
|
||||
|
||||
# check for correct behavior whether or not data is 1d to begin with
|
||||
with pytest.raises(ValueError, match=message):
|
||||
hypotest(*data, axis=None, nan_policy=nan_policy,
|
||||
*args, **kwds)
|
||||
with pytest.raises(ValueError, match=message):
|
||||
hypotest(*data_raveled, axis=None, nan_policy=nan_policy,
|
||||
*args, **kwds)
|
||||
|
||||
else:
|
||||
# behavior of reference implementation with 1d input, hypotest with 1d
|
||||
# input, and hypotest with Nd input should match, whether that means
|
||||
# that outputs are equal or they raise the same exception
|
||||
|
||||
ea_str, eb_str, ec_str = None, None, None
|
||||
with np.errstate(divide='ignore', invalid='ignore'):
|
||||
try:
|
||||
res1da = nan_policy_1d(hypotest, data_raveled, unpacker, *args,
|
||||
nan_policy=nan_policy, paired=paired,
|
||||
_no_deco=True, **kwds)
|
||||
except (RuntimeWarning, ValueError, ZeroDivisionError) as ea:
|
||||
ea_str = str(ea)
|
||||
|
||||
try:
|
||||
res1db = unpacker(hypotest(*data_raveled, *args,
|
||||
nan_policy=nan_policy, **kwds))
|
||||
except (RuntimeWarning, ValueError, ZeroDivisionError) as eb:
|
||||
eb_str = str(eb)
|
||||
|
||||
try:
|
||||
res1dc = unpacker(hypotest(*data, *args, axis=None,
|
||||
nan_policy=nan_policy, **kwds))
|
||||
except (RuntimeWarning, ValueError, ZeroDivisionError) as ec:
|
||||
ec_str = str(ec)
|
||||
|
||||
if ea_str or eb_str or ec_str:
|
||||
assert any([str(ea_str).startswith(message)
|
||||
for message in too_small_messages])
|
||||
assert ea_str == eb_str == ec_str
|
||||
else:
|
||||
assert_equal(res1db, res1da)
|
||||
assert_equal(res1dc, res1da)
|
||||
|
||||
|
||||
@pytest.mark.parametrize(("axis"), (0, 1, 2))
|
||||
def test_axis_nan_policy_decorated_positional_axis(axis):
|
||||
# Test for correct behavior of function decorated with
|
||||
# _axis_nan_policy_decorator whether `axis` is provided as positional or
|
||||
# keyword argument
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
pytest.xfail("Avoid test failures due to old version of NumPy")
|
||||
|
||||
shape = (8, 9, 10)
|
||||
rng = np.random.default_rng(0)
|
||||
x = rng.random(shape)
|
||||
y = rng.random(shape)
|
||||
res1 = stats.mannwhitneyu(x, y, True, 'two-sided', axis)
|
||||
res2 = stats.mannwhitneyu(x, y, True, 'two-sided', axis=axis)
|
||||
assert_equal(res1, res2)
|
||||
|
||||
message = "mannwhitneyu() got multiple values for argument 'axis'"
|
||||
with pytest.raises(TypeError, match=re.escape(message)):
|
||||
stats.mannwhitneyu(x, y, True, 'two-sided', axis, axis=axis)
|
||||
|
||||
|
||||
def test_axis_nan_policy_decorated_positional_args():
|
||||
# Test for correct behavior of function decorated with
|
||||
# _axis_nan_policy_decorator when function accepts *args
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
pytest.xfail("Avoid test failures due to old version of NumPy")
|
||||
|
||||
shape = (3, 8, 9, 10)
|
||||
rng = np.random.default_rng(0)
|
||||
x = rng.random(shape)
|
||||
x[0, 0, 0, 0] = np.nan
|
||||
stats.kruskal(*x)
|
||||
|
||||
message = "kruskal() got an unexpected keyword argument 'args'"
|
||||
with pytest.raises(TypeError, match=re.escape(message)):
|
||||
stats.kruskal(args=x)
|
||||
|
||||
with pytest.raises(TypeError, match=re.escape(message)):
|
||||
stats.kruskal(*x, args=x)
|
||||
|
||||
|
||||
def test_axis_nan_policy_decorated_keyword_samples():
|
||||
# Test for correct behavior of function decorated with
|
||||
# _axis_nan_policy_decorator whether samples are provided as positional or
|
||||
# keyword arguments
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
pytest.xfail("Avoid test failures due to old version of NumPy")
|
||||
|
||||
shape = (2, 8, 9, 10)
|
||||
rng = np.random.default_rng(0)
|
||||
x = rng.random(shape)
|
||||
x[0, 0, 0, 0] = np.nan
|
||||
res1 = stats.mannwhitneyu(*x)
|
||||
res2 = stats.mannwhitneyu(x=x[0], y=x[1])
|
||||
assert_equal(res1, res2)
|
||||
|
||||
message = "mannwhitneyu() got multiple values for argument"
|
||||
with pytest.raises(TypeError, match=re.escape(message)):
|
||||
stats.mannwhitneyu(*x, x=x[0], y=x[1])
|
||||
|
||||
|
||||
@pytest.mark.parametrize(("hypotest", "args", "kwds", "n_samples", "paired",
|
||||
"unpacker"), axis_nan_policy_cases)
|
||||
def test_axis_nan_policy_decorated_pickled(hypotest, args, kwds, n_samples,
|
||||
paired, unpacker):
|
||||
if NumpyVersion(np.__version__) < '1.18.0':
|
||||
rng = np.random.RandomState(0)
|
||||
else:
|
||||
rng = np.random.default_rng(0)
|
||||
|
||||
# Some hypothesis tests return a non-iterable that needs an `unpacker` to
|
||||
# extract the statistic and p-value. For those that don't:
|
||||
if not unpacker:
|
||||
def unpacker(res):
|
||||
return res
|
||||
|
||||
data = rng.uniform(size=(n_samples, 2, 30))
|
||||
pickled_hypotest = pickle.dumps(hypotest)
|
||||
unpickled_hypotest = pickle.loads(pickled_hypotest)
|
||||
res1 = unpacker(hypotest(*data, *args, axis=-1, **kwds))
|
||||
res2 = unpacker(unpickled_hypotest(*data, *args, axis=-1, **kwds))
|
||||
assert_allclose(res1, res2, rtol=1e-12)
|
||||
|
||||
|
||||
def test_check_empty_inputs():
|
||||
# Test that _check_empty_inputs is doing its job, at least for single-
|
||||
# sample inputs. (Multi-sample functionality is tested below.)
|
||||
# If the input sample is not empty, it should return None.
|
||||
# If the input sample is empty, it should return an array of NaNs or an
|
||||
# empty array of appropriate shape. np.mean is used as a reference for the
|
||||
# output because, like the statistics calculated by these functions,
|
||||
# it works along and "consumes" `axis` but preserves the other axes.
|
||||
for i in range(5):
|
||||
for combo in combinations_with_replacement([0, 1, 2], i):
|
||||
for axis in range(len(combo)):
|
||||
samples = (np.zeros(combo),)
|
||||
output = stats._axis_nan_policy._check_empty_inputs(samples,
|
||||
axis)
|
||||
if output is not None:
|
||||
with np.testing.suppress_warnings() as sup:
|
||||
sup.filter(RuntimeWarning, "Mean of empty slice.")
|
||||
sup.filter(RuntimeWarning, "invalid value encountered")
|
||||
reference = samples[0].mean(axis=axis)
|
||||
np.testing.assert_equal(output, reference)
|
||||
|
||||
|
||||
def _check_arrays_broadcastable(arrays, axis):
|
||||
# https://numpy.org/doc/stable/user/basics.broadcasting.html
|
||||
# "When operating on two arrays, NumPy compares their shapes element-wise.
|
||||
# It starts with the trailing (i.e. rightmost) dimensions and works its
|
||||
# way left.
|
||||
# Two dimensions are compatible when
|
||||
# 1. they are equal, or
|
||||
# 2. one of them is 1
|
||||
# ...
|
||||
# Arrays do not need to have the same number of dimensions."
|
||||
# (Clarification: if the arrays are compatible according to the criteria
|
||||
# above and an array runs out of dimensions, it is still compatible.)
|
||||
# Below, we follow the rules above except ignoring `axis`
|
||||
|
||||
n_dims = max([arr.ndim for arr in arrays])
|
||||
if axis is not None:
|
||||
# convert to negative axis
|
||||
axis = (-n_dims + axis) if axis >= 0 else axis
|
||||
|
||||
for dim in range(1, n_dims+1): # we'll index from -1 to -n_dims, inclusive
|
||||
if -dim == axis:
|
||||
continue # ignore lengths along `axis`
|
||||
|
||||
dim_lengths = set()
|
||||
for arr in arrays:
|
||||
if dim <= arr.ndim and arr.shape[-dim] != 1:
|
||||
dim_lengths.add(arr.shape[-dim])
|
||||
|
||||
if len(dim_lengths) > 1:
|
||||
return False
|
||||
return True
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
@pytest.mark.parametrize(("hypotest", "args", "kwds", "n_samples", "paired",
|
||||
"unpacker"), axis_nan_policy_cases)
|
||||
def test_empty(hypotest, args, kwds, n_samples, paired, unpacker):
|
||||
# test for correct output shape when at least one input is empty
|
||||
|
||||
def small_data_generator(n_samples, n_dims):
|
||||
|
||||
def small_sample_generator(n_dims):
|
||||
# return all possible "small" arrays in up to n_dim dimensions
|
||||
for i in n_dims:
|
||||
# "small" means with size along dimension either 0 or 1
|
||||
for combo in combinations_with_replacement([0, 1, 2], i):
|
||||
yield np.zeros(combo)
|
||||
|
||||
# yield all possible combinations of small samples
|
||||
gens = [small_sample_generator(n_dims) for i in range(n_samples)]
|
||||
for i in product(*gens):
|
||||
yield i
|
||||
|
||||
n_dims = [2, 3]
|
||||
for samples in small_data_generator(n_samples, n_dims):
|
||||
|
||||
# this test is only for arrays of zero size
|
||||
if not any((sample.size == 0 for sample in samples)):
|
||||
continue
|
||||
|
||||
max_axis = max((sample.ndim for sample in samples))
|
||||
|
||||
# need to test for all valid values of `axis` parameter, too
|
||||
for axis in range(-max_axis, max_axis):
|
||||
|
||||
try:
|
||||
# After broadcasting, all arrays are the same shape, so
|
||||
# the shape of the output should be the same as a single-
|
||||
# sample statistic. Use np.mean as a reference.
|
||||
concat = stats._stats_py._broadcast_concatenate(samples, axis)
|
||||
with np.testing.suppress_warnings() as sup:
|
||||
sup.filter(RuntimeWarning, "Mean of empty slice.")
|
||||
sup.filter(RuntimeWarning, "invalid value encountered")
|
||||
expected = np.mean(concat, axis=axis) * np.nan
|
||||
|
||||
res = hypotest(*samples, *args, axis=axis, **kwds)
|
||||
|
||||
if hasattr(res, 'statistic'):
|
||||
assert_equal(res.statistic, expected)
|
||||
assert_equal(res.pvalue, expected)
|
||||
else:
|
||||
assert_equal(res, expected)
|
||||
|
||||
except ValueError:
|
||||
# confirm that the arrays truly are not broadcastable
|
||||
assert not _check_arrays_broadcastable(samples, axis)
|
||||
|
||||
# confirm that _both_ `_broadcast_concatenate` and `hypotest`
|
||||
# produce this information.
|
||||
message = "Array shapes are incompatible for broadcasting."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
stats._stats_py._broadcast_concatenate(samples, axis)
|
||||
with pytest.raises(ValueError, match=message):
|
||||
hypotest(*samples, *args, axis=axis, **kwds)
|
||||
@@ -0,0 +1,542 @@
|
||||
import numpy as np
|
||||
from numpy.testing import assert_allclose
|
||||
from pytest import raises as assert_raises
|
||||
from scipy.stats import (binned_statistic, binned_statistic_2d,
|
||||
binned_statistic_dd)
|
||||
from scipy._lib._util import check_random_state
|
||||
|
||||
from .common_tests import check_named_results
|
||||
|
||||
|
||||
class TestBinnedStatistic:
|
||||
|
||||
@classmethod
|
||||
def setup_class(cls):
|
||||
rng = check_random_state(9865)
|
||||
cls.x = rng.uniform(size=100)
|
||||
cls.y = rng.uniform(size=100)
|
||||
cls.v = rng.uniform(size=100)
|
||||
cls.X = rng.uniform(size=(100, 3))
|
||||
cls.w = rng.uniform(size=100)
|
||||
cls.u = rng.uniform(size=100) + 1e6
|
||||
|
||||
def test_1d_count(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
count1, edges1, bc = binned_statistic(x, v, 'count', bins=10)
|
||||
count2, edges2 = np.histogram(x, bins=10)
|
||||
|
||||
assert_allclose(count1, count2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_gh5927(self):
|
||||
# smoke test for gh5927 - binned_statistic was using `is` for string
|
||||
# comparison
|
||||
x = self.x
|
||||
v = self.v
|
||||
statistics = [u'mean', u'median', u'count', u'sum']
|
||||
for statistic in statistics:
|
||||
binned_statistic(x, v, statistic, bins=10)
|
||||
|
||||
def test_big_number_std(self):
|
||||
# tests for numerical stability of std calculation
|
||||
# see issue gh-10126 for more
|
||||
x = self.x
|
||||
u = self.u
|
||||
stat1, edges1, bc = binned_statistic(x, u, 'std', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, u, np.std, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
|
||||
def test_empty_bins_std(self):
|
||||
# tests that std returns gives nan for empty bins
|
||||
x = self.x
|
||||
u = self.u
|
||||
print(binned_statistic(x, u, 'count', bins=1000))
|
||||
stat1, edges1, bc = binned_statistic(x, u, 'std', bins=1000)
|
||||
stat2, edges2, bc = binned_statistic(x, u, np.std, bins=1000)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
|
||||
def test_non_finite_inputs_and_int_bins(self):
|
||||
# if either `values` or `sample` contain np.inf or np.nan throw
|
||||
# see issue gh-9010 for more
|
||||
x = self.x
|
||||
u = self.u
|
||||
orig = u[0]
|
||||
u[0] = np.inf
|
||||
assert_raises(ValueError, binned_statistic, u, x, 'std', bins=10)
|
||||
# need to test for non-python specific ints, e.g. np.int8, np.int64
|
||||
assert_raises(ValueError, binned_statistic, u, x, 'std',
|
||||
bins=np.int64(10))
|
||||
u[0] = np.nan
|
||||
assert_raises(ValueError, binned_statistic, u, x, 'count', bins=10)
|
||||
# replace original value, u belongs the class
|
||||
u[0] = orig
|
||||
|
||||
def test_1d_result_attributes(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
res = binned_statistic(x, v, 'count', bins=10)
|
||||
attributes = ('statistic', 'bin_edges', 'binnumber')
|
||||
check_named_results(res, attributes)
|
||||
|
||||
def test_1d_sum(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
sum1, edges1, bc = binned_statistic(x, v, 'sum', bins=10)
|
||||
sum2, edges2 = np.histogram(x, bins=10, weights=v)
|
||||
|
||||
assert_allclose(sum1, sum2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_mean(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic(x, v, 'mean', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, v, np.mean, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_std(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic(x, v, 'std', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, v, np.std, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_min(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic(x, v, 'min', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, v, np.min, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_max(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic(x, v, 'max', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, v, np.max, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_median(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic(x, v, 'median', bins=10)
|
||||
stat2, edges2, bc = binned_statistic(x, v, np.median, bins=10)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_1d_bincode(self):
|
||||
x = self.x[:20]
|
||||
v = self.v[:20]
|
||||
|
||||
count1, edges1, bc = binned_statistic(x, v, 'count', bins=3)
|
||||
bc2 = np.array([3, 2, 1, 3, 2, 3, 3, 3, 3, 1, 1, 3, 3, 1, 2, 3, 1,
|
||||
1, 2, 1])
|
||||
|
||||
bcount = [(bc == i).sum() for i in np.unique(bc)]
|
||||
|
||||
assert_allclose(bc, bc2)
|
||||
assert_allclose(bcount, count1)
|
||||
|
||||
def test_1d_range_keyword(self):
|
||||
# Regression test for gh-3063, range can be (min, max) or [(min, max)]
|
||||
np.random.seed(9865)
|
||||
x = np.arange(30)
|
||||
data = np.random.random(30)
|
||||
|
||||
mean, bins, _ = binned_statistic(x[:15], data[:15])
|
||||
mean_range, bins_range, _ = binned_statistic(x, data, range=[(0, 14)])
|
||||
mean_range2, bins_range2, _ = binned_statistic(x, data, range=(0, 14))
|
||||
|
||||
assert_allclose(mean, mean_range)
|
||||
assert_allclose(bins, bins_range)
|
||||
assert_allclose(mean, mean_range2)
|
||||
assert_allclose(bins, bins_range2)
|
||||
|
||||
def test_1d_multi_values(self):
|
||||
x = self.x
|
||||
v = self.v
|
||||
w = self.w
|
||||
|
||||
stat1v, edges1v, bc1v = binned_statistic(x, v, 'mean', bins=10)
|
||||
stat1w, edges1w, bc1w = binned_statistic(x, w, 'mean', bins=10)
|
||||
stat2, edges2, bc2 = binned_statistic(x, [v, w], 'mean', bins=10)
|
||||
|
||||
assert_allclose(stat2[0], stat1v)
|
||||
assert_allclose(stat2[1], stat1w)
|
||||
assert_allclose(edges1v, edges2)
|
||||
assert_allclose(bc1v, bc2)
|
||||
|
||||
def test_2d_count(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
count1, binx1, biny1, bc = binned_statistic_2d(
|
||||
x, y, v, 'count', bins=5)
|
||||
count2, binx2, biny2 = np.histogram2d(x, y, bins=5)
|
||||
|
||||
assert_allclose(count1, count2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_result_attributes(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
res = binned_statistic_2d(x, y, v, 'count', bins=5)
|
||||
attributes = ('statistic', 'x_edge', 'y_edge', 'binnumber')
|
||||
check_named_results(res, attributes)
|
||||
|
||||
def test_2d_sum(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
sum1, binx1, biny1, bc = binned_statistic_2d(x, y, v, 'sum', bins=5)
|
||||
sum2, binx2, biny2 = np.histogram2d(x, y, bins=5, weights=v)
|
||||
|
||||
assert_allclose(sum1, sum2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_mean(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(x, y, v, 'mean', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(x, y, v, np.mean, bins=5)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_mean_unicode(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(
|
||||
x, y, v, 'mean', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(x, y, v, np.mean, bins=5)
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_std(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(x, y, v, 'std', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(x, y, v, np.std, bins=5)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_min(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(x, y, v, 'min', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(x, y, v, np.min, bins=5)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_max(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(x, y, v, 'max', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(x, y, v, np.max, bins=5)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_median(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat1, binx1, biny1, bc = binned_statistic_2d(
|
||||
x, y, v, 'median', bins=5)
|
||||
stat2, binx2, biny2, bc = binned_statistic_2d(
|
||||
x, y, v, np.median, bins=5)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(binx1, binx2)
|
||||
assert_allclose(biny1, biny2)
|
||||
|
||||
def test_2d_bincode(self):
|
||||
x = self.x[:20]
|
||||
y = self.y[:20]
|
||||
v = self.v[:20]
|
||||
|
||||
count1, binx1, biny1, bc = binned_statistic_2d(
|
||||
x, y, v, 'count', bins=3)
|
||||
bc2 = np.array([17, 11, 6, 16, 11, 17, 18, 17, 17, 7, 6, 18, 16,
|
||||
6, 11, 16, 6, 6, 11, 8])
|
||||
|
||||
bcount = [(bc == i).sum() for i in np.unique(bc)]
|
||||
|
||||
assert_allclose(bc, bc2)
|
||||
count1adj = count1[count1.nonzero()]
|
||||
assert_allclose(bcount, count1adj)
|
||||
|
||||
def test_2d_multi_values(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
w = self.w
|
||||
|
||||
stat1v, binx1v, biny1v, bc1v = binned_statistic_2d(
|
||||
x, y, v, 'mean', bins=8)
|
||||
stat1w, binx1w, biny1w, bc1w = binned_statistic_2d(
|
||||
x, y, w, 'mean', bins=8)
|
||||
stat2, binx2, biny2, bc2 = binned_statistic_2d(
|
||||
x, y, [v, w], 'mean', bins=8)
|
||||
|
||||
assert_allclose(stat2[0], stat1v)
|
||||
assert_allclose(stat2[1], stat1w)
|
||||
assert_allclose(binx1v, binx2)
|
||||
assert_allclose(biny1w, biny2)
|
||||
assert_allclose(bc1v, bc2)
|
||||
|
||||
def test_2d_binnumbers_unraveled(self):
|
||||
x = self.x
|
||||
y = self.y
|
||||
v = self.v
|
||||
|
||||
stat, edgesx, bcx = binned_statistic(x, v, 'mean', bins=20)
|
||||
stat, edgesy, bcy = binned_statistic(y, v, 'mean', bins=10)
|
||||
|
||||
stat2, edgesx2, edgesy2, bc2 = binned_statistic_2d(
|
||||
x, y, v, 'mean', bins=(20, 10), expand_binnumbers=True)
|
||||
|
||||
bcx3 = np.searchsorted(edgesx, x, side='right')
|
||||
bcy3 = np.searchsorted(edgesy, y, side='right')
|
||||
|
||||
# `numpy.searchsorted` is non-inclusive on right-edge, compensate
|
||||
bcx3[x == x.max()] -= 1
|
||||
bcy3[y == y.max()] -= 1
|
||||
|
||||
assert_allclose(bcx, bc2[0])
|
||||
assert_allclose(bcy, bc2[1])
|
||||
assert_allclose(bcx3, bc2[0])
|
||||
assert_allclose(bcy3, bc2[1])
|
||||
|
||||
def test_dd_count(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
count1, edges1, bc = binned_statistic_dd(X, v, 'count', bins=3)
|
||||
count2, edges2 = np.histogramdd(X, bins=3)
|
||||
|
||||
assert_allclose(count1, count2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_result_attributes(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
res = binned_statistic_dd(X, v, 'count', bins=3)
|
||||
attributes = ('statistic', 'bin_edges', 'binnumber')
|
||||
check_named_results(res, attributes)
|
||||
|
||||
def test_dd_sum(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
sum1, edges1, bc = binned_statistic_dd(X, v, 'sum', bins=3)
|
||||
sum2, edges2 = np.histogramdd(X, bins=3, weights=v)
|
||||
|
||||
assert_allclose(sum1, sum2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_mean(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic_dd(X, v, 'mean', bins=3)
|
||||
stat2, edges2, bc = binned_statistic_dd(X, v, np.mean, bins=3)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_std(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic_dd(X, v, 'std', bins=3)
|
||||
stat2, edges2, bc = binned_statistic_dd(X, v, np.std, bins=3)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_min(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic_dd(X, v, 'min', bins=3)
|
||||
stat2, edges2, bc = binned_statistic_dd(X, v, np.min, bins=3)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_max(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic_dd(X, v, 'max', bins=3)
|
||||
stat2, edges2, bc = binned_statistic_dd(X, v, np.max, bins=3)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_median(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat1, edges1, bc = binned_statistic_dd(X, v, 'median', bins=3)
|
||||
stat2, edges2, bc = binned_statistic_dd(X, v, np.median, bins=3)
|
||||
|
||||
assert_allclose(stat1, stat2)
|
||||
assert_allclose(edges1, edges2)
|
||||
|
||||
def test_dd_bincode(self):
|
||||
X = self.X[:20]
|
||||
v = self.v[:20]
|
||||
|
||||
count1, edges1, bc = binned_statistic_dd(X, v, 'count', bins=3)
|
||||
bc2 = np.array([63, 33, 86, 83, 88, 67, 57, 33, 42, 41, 82, 83, 92,
|
||||
32, 36, 91, 43, 87, 81, 81])
|
||||
|
||||
bcount = [(bc == i).sum() for i in np.unique(bc)]
|
||||
|
||||
assert_allclose(bc, bc2)
|
||||
count1adj = count1[count1.nonzero()]
|
||||
assert_allclose(bcount, count1adj)
|
||||
|
||||
def test_dd_multi_values(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
w = self.w
|
||||
|
||||
for stat in ["count", "sum", "mean", "std", "min", "max", "median",
|
||||
np.std]:
|
||||
stat1v, edges1v, bc1v = binned_statistic_dd(X, v, stat, bins=8)
|
||||
stat1w, edges1w, bc1w = binned_statistic_dd(X, w, stat, bins=8)
|
||||
stat2, edges2, bc2 = binned_statistic_dd(X, [v, w], stat, bins=8)
|
||||
assert_allclose(stat2[0], stat1v)
|
||||
assert_allclose(stat2[1], stat1w)
|
||||
assert_allclose(edges1v, edges2)
|
||||
assert_allclose(edges1w, edges2)
|
||||
assert_allclose(bc1v, bc2)
|
||||
|
||||
def test_dd_binnumbers_unraveled(self):
|
||||
X = self.X
|
||||
v = self.v
|
||||
|
||||
stat, edgesx, bcx = binned_statistic(X[:, 0], v, 'mean', bins=15)
|
||||
stat, edgesy, bcy = binned_statistic(X[:, 1], v, 'mean', bins=20)
|
||||
stat, edgesz, bcz = binned_statistic(X[:, 2], v, 'mean', bins=10)
|
||||
|
||||
stat2, edges2, bc2 = binned_statistic_dd(
|
||||
X, v, 'mean', bins=(15, 20, 10), expand_binnumbers=True)
|
||||
|
||||
assert_allclose(bcx, bc2[0])
|
||||
assert_allclose(bcy, bc2[1])
|
||||
assert_allclose(bcz, bc2[2])
|
||||
|
||||
def test_dd_binned_statistic_result(self):
|
||||
# NOTE: tests the reuse of bin_edges from previous call
|
||||
x = np.random.random((10000, 3))
|
||||
v = np.random.random((10000))
|
||||
bins = np.linspace(0, 1, 10)
|
||||
bins = (bins, bins, bins)
|
||||
|
||||
result = binned_statistic_dd(x, v, 'mean', bins=bins)
|
||||
stat = result.statistic
|
||||
|
||||
result = binned_statistic_dd(x, v, 'mean',
|
||||
binned_statistic_result=result)
|
||||
stat2 = result.statistic
|
||||
|
||||
assert_allclose(stat, stat2)
|
||||
|
||||
def test_dd_zero_dedges(self):
|
||||
x = np.random.random((10000, 3))
|
||||
v = np.random.random((10000))
|
||||
bins = np.linspace(0, 1, 10)
|
||||
bins = np.append(bins, 1)
|
||||
bins = (bins, bins, bins)
|
||||
with assert_raises(ValueError, match='difference is numerically 0'):
|
||||
binned_statistic_dd(x, v, 'mean', bins=bins)
|
||||
|
||||
def test_dd_range_errors(self):
|
||||
# Test that descriptive exceptions are raised as appropriate for bad
|
||||
# values of the `range` argument. (See gh-12996)
|
||||
with assert_raises(ValueError,
|
||||
match='In range, start must be <= stop'):
|
||||
binned_statistic_dd([self.y], self.v,
|
||||
range=[[1, 0]])
|
||||
with assert_raises(
|
||||
ValueError,
|
||||
match='In dimension 1 of range, start must be <= stop'):
|
||||
binned_statistic_dd([self.x, self.y], self.v,
|
||||
range=[[1, 0], [0, 1]])
|
||||
with assert_raises(
|
||||
ValueError,
|
||||
match='In dimension 2 of range, start must be <= stop'):
|
||||
binned_statistic_dd([self.x, self.y], self.v,
|
||||
range=[[0, 1], [1, 0]])
|
||||
with assert_raises(
|
||||
ValueError,
|
||||
match='range given for 1 dimensions; 2 required'):
|
||||
binned_statistic_dd([self.x, self.y], self.v,
|
||||
range=[[0, 1]])
|
||||
|
||||
def test_binned_statistic_float32(self):
|
||||
X = np.array([0, 0.42358226], dtype=np.float32)
|
||||
stat, _, _ = binned_statistic(X, None, 'count', bins=5)
|
||||
assert_allclose(stat, np.array([1, 0, 0, 0, 1], dtype=np.float64))
|
||||
|
||||
def test_gh14332(self):
|
||||
# Test the wrong output when the `sample` is close to bin edge
|
||||
x = []
|
||||
size = 20
|
||||
for i in range(size):
|
||||
x += [1-0.1**i]
|
||||
|
||||
bins = np.linspace(0,1,11)
|
||||
sum1, edges1, bc = binned_statistic_dd(x, np.ones(len(x)),
|
||||
bins=[bins], statistic='sum')
|
||||
sum2, edges2 = np.histogram(x, bins=bins)
|
||||
|
||||
assert_allclose(sum1, sum2)
|
||||
assert_allclose(edges1[0], edges2)
|
||||
@@ -0,0 +1,490 @@
|
||||
import numpy as np
|
||||
import pytest
|
||||
from scipy.stats import bootstrap, BootstrapDegenerateDistributionWarning
|
||||
from numpy.testing import assert_allclose, assert_equal
|
||||
from scipy import stats
|
||||
from .. import _bootstrap as _bootstrap
|
||||
from scipy._lib._util import rng_integers
|
||||
|
||||
|
||||
def test_bootstrap_iv():
|
||||
|
||||
message = "`data` must be a sequence of samples."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(1, np.mean)
|
||||
|
||||
message = "`data` must contain at least one sample."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(tuple(), np.mean)
|
||||
|
||||
message = "each sample in `data` must contain two or more observations..."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3], [1]), np.mean)
|
||||
|
||||
message = ("When `paired is True`, all samples must have the same length ")
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3], [1, 2, 3, 4]), np.mean, paired=True)
|
||||
|
||||
message = "`vectorized` must be `True` or `False`."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(1, np.mean, vectorized='ekki')
|
||||
|
||||
message = "`axis` must be an integer."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, axis=1.5)
|
||||
|
||||
message = "could not convert string to float"
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, confidence_level='ni')
|
||||
|
||||
message = "`n_resamples` must be a positive integer."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, n_resamples=-1000)
|
||||
|
||||
message = "`n_resamples` must be a positive integer."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, n_resamples=1000.5)
|
||||
|
||||
message = "`batch` must be a positive integer or None."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, batch=-1000)
|
||||
|
||||
message = "`batch` must be a positive integer or None."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, batch=1000.5)
|
||||
|
||||
message = "`method` must be in"
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, method='ekki')
|
||||
|
||||
message = "`method = 'BCa' is only available for one-sample statistics"
|
||||
|
||||
def statistic(x, y, axis):
|
||||
mean1 = np.mean(x, axis)
|
||||
mean2 = np.mean(y, axis)
|
||||
return mean1 - mean2
|
||||
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([.1, .2, .3], [.1, .2, .3]), statistic, method='BCa')
|
||||
|
||||
message = "'herring' cannot be used to seed a"
|
||||
with pytest.raises(ValueError, match=message):
|
||||
bootstrap(([1, 2, 3],), np.mean, random_state='herring')
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ['basic', 'percentile', 'BCa'])
|
||||
@pytest.mark.parametrize("axis", [0, 1, 2])
|
||||
def test_bootstrap_batch(method, axis):
|
||||
# for one-sample statistics, batch size shouldn't affect the result
|
||||
np.random.seed(0)
|
||||
|
||||
x = np.random.rand(10, 11, 12)
|
||||
res1 = bootstrap((x,), np.mean, batch=None, method=method,
|
||||
random_state=0, axis=axis, n_resamples=100)
|
||||
res2 = bootstrap((x,), np.mean, batch=10, method=method,
|
||||
random_state=0, axis=axis, n_resamples=100)
|
||||
|
||||
assert_equal(res2.confidence_interval.low, res1.confidence_interval.low)
|
||||
assert_equal(res2.confidence_interval.high, res1.confidence_interval.high)
|
||||
assert_equal(res2.standard_error, res1.standard_error)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ['basic', 'percentile', 'BCa'])
|
||||
def test_bootstrap_paired(method):
|
||||
# test that `paired` works as expected
|
||||
np.random.seed(0)
|
||||
n = 100
|
||||
x = np.random.rand(n)
|
||||
y = np.random.rand(n)
|
||||
|
||||
def my_statistic(x, y, axis=-1):
|
||||
return ((x-y)**2).mean(axis=axis)
|
||||
|
||||
def my_paired_statistic(i, axis=-1):
|
||||
a = x[i]
|
||||
b = y[i]
|
||||
res = my_statistic(a, b)
|
||||
return res
|
||||
|
||||
i = np.arange(len(x))
|
||||
|
||||
res1 = bootstrap((i,), my_paired_statistic, random_state=0)
|
||||
res2 = bootstrap((x, y), my_statistic, paired=True, random_state=0)
|
||||
|
||||
assert_allclose(res1.confidence_interval, res2.confidence_interval)
|
||||
assert_allclose(res1.standard_error, res2.standard_error)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ['basic', 'percentile', 'BCa'])
|
||||
@pytest.mark.parametrize("axis", [0, 1, 2])
|
||||
@pytest.mark.parametrize("paired", [True, False])
|
||||
def test_bootstrap_vectorized(method, axis, paired):
|
||||
# test that paired is vectorized as expected: when samples are tiled,
|
||||
# CI and standard_error of each axis-slice is the same as those of the
|
||||
# original 1d sample
|
||||
|
||||
if not paired and method == 'BCa':
|
||||
# should re-assess when BCa is extended
|
||||
pytest.xfail(reason="BCa currently for 1-sample statistics only")
|
||||
np.random.seed(0)
|
||||
|
||||
def my_statistic(x, y, z, axis=-1):
|
||||
return x.mean(axis=axis) + y.mean(axis=axis) + z.mean(axis=axis)
|
||||
|
||||
shape = 10, 11, 12
|
||||
n_samples = shape[axis]
|
||||
|
||||
x = np.random.rand(n_samples)
|
||||
y = np.random.rand(n_samples)
|
||||
z = np.random.rand(n_samples)
|
||||
res1 = bootstrap((x, y, z), my_statistic, paired=paired, method=method,
|
||||
random_state=0, axis=0, n_resamples=100)
|
||||
|
||||
reshape = [1, 1, 1]
|
||||
reshape[axis] = n_samples
|
||||
x = np.broadcast_to(x.reshape(reshape), shape)
|
||||
y = np.broadcast_to(y.reshape(reshape), shape)
|
||||
z = np.broadcast_to(z.reshape(reshape), shape)
|
||||
res2 = bootstrap((x, y, z), my_statistic, paired=paired, method=method,
|
||||
random_state=0, axis=axis, n_resamples=100)
|
||||
|
||||
assert_allclose(res2.confidence_interval.low,
|
||||
res1.confidence_interval.low)
|
||||
assert_allclose(res2.confidence_interval.high,
|
||||
res1.confidence_interval.high)
|
||||
assert_allclose(res2.standard_error, res1.standard_error)
|
||||
|
||||
result_shape = list(shape)
|
||||
result_shape.pop(axis)
|
||||
|
||||
assert_equal(res2.confidence_interval.low.shape, result_shape)
|
||||
assert_equal(res2.confidence_interval.high.shape, result_shape)
|
||||
assert_equal(res2.standard_error.shape, result_shape)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ['basic', 'percentile', 'BCa'])
|
||||
def test_bootstrap_against_theory(method):
|
||||
# based on https://www.statology.org/confidence-intervals-python/
|
||||
data = stats.norm.rvs(loc=5, scale=2, size=5000, random_state=0)
|
||||
alpha = 0.95
|
||||
dist = stats.t(df=len(data)-1, loc=np.mean(data), scale=stats.sem(data))
|
||||
expected_interval = dist.interval(alpha=alpha)
|
||||
expected_se = dist.std()
|
||||
|
||||
res = bootstrap((data,), np.mean, n_resamples=5000,
|
||||
confidence_level=alpha, method=method,
|
||||
random_state=0)
|
||||
assert_allclose(res.confidence_interval, expected_interval, rtol=5e-4)
|
||||
assert_allclose(res.standard_error, expected_se, atol=3e-4)
|
||||
|
||||
|
||||
tests_R = {"basic": (23.77, 79.12),
|
||||
"percentile": (28.86, 84.21),
|
||||
"BCa": (32.31, 91.43)}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method, expected", tests_R.items())
|
||||
def test_bootstrap_against_R(method, expected):
|
||||
# Compare against R's "boot" library
|
||||
# library(boot)
|
||||
|
||||
# stat <- function (x, a) {
|
||||
# mean(x[a])
|
||||
# }
|
||||
|
||||
# x <- c(10, 12, 12.5, 12.5, 13.9, 15, 21, 22,
|
||||
# 23, 34, 50, 81, 89, 121, 134, 213)
|
||||
|
||||
# # Use a large value so we get a few significant digits for the CI.
|
||||
# n = 1000000
|
||||
# bootresult = boot(x, stat, n)
|
||||
# result <- boot.ci(bootresult)
|
||||
# print(result)
|
||||
x = np.array([10, 12, 12.5, 12.5, 13.9, 15, 21, 22,
|
||||
23, 34, 50, 81, 89, 121, 134, 213])
|
||||
res = bootstrap((x,), np.mean, n_resamples=1000000, method=method,
|
||||
random_state=0)
|
||||
assert_allclose(res.confidence_interval, expected, rtol=0.005)
|
||||
|
||||
|
||||
tests_against_itself_1samp = {"basic": 1780,
|
||||
"percentile": 1784,
|
||||
"BCa": 1784}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method, expected",
|
||||
tests_against_itself_1samp.items())
|
||||
def test_bootstrap_against_itself_1samp(method, expected):
|
||||
# The expected values in this test were generated using bootstrap
|
||||
# to check for unintended changes in behavior. The test also makes sure
|
||||
# that bootstrap works with multi-sample statistics and that the
|
||||
# `axis` argument works as expected / function is vectorized.
|
||||
np.random.seed(0)
|
||||
|
||||
n = 100 # size of sample
|
||||
n_resamples = 999 # number of bootstrap resamples used to form each CI
|
||||
confidence_level = 0.9
|
||||
|
||||
# The true mean is 5
|
||||
dist = stats.norm(loc=5, scale=1)
|
||||
stat_true = dist.mean()
|
||||
|
||||
# Do the same thing 2000 times. (The code is fully vectorized.)
|
||||
n_replications = 2000
|
||||
data = dist.rvs(size=(n_replications, n))
|
||||
res = bootstrap((data,),
|
||||
statistic=np.mean,
|
||||
confidence_level=confidence_level,
|
||||
n_resamples=n_resamples,
|
||||
batch=50,
|
||||
method=method,
|
||||
axis=-1)
|
||||
ci = res.confidence_interval
|
||||
|
||||
# ci contains vectors of lower and upper confidence interval bounds
|
||||
ci_contains_true = np.sum((ci[0] < stat_true) & (stat_true < ci[1]))
|
||||
assert ci_contains_true == expected
|
||||
|
||||
# ci_contains_true is not inconsistent with confidence_level
|
||||
pvalue = stats.binomtest(ci_contains_true, n_replications,
|
||||
confidence_level).pvalue
|
||||
assert pvalue > 0.1
|
||||
|
||||
|
||||
tests_against_itself_2samp = {"basic": 892,
|
||||
"percentile": 890}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method, expected",
|
||||
tests_against_itself_2samp.items())
|
||||
def test_bootstrap_against_itself_2samp(method, expected):
|
||||
# The expected values in this test were generated using bootstrap
|
||||
# to check for unintended changes in behavior. The test also makes sure
|
||||
# that bootstrap works with multi-sample statistics and that the
|
||||
# `axis` argument works as expected / function is vectorized.
|
||||
np.random.seed(0)
|
||||
|
||||
n1 = 100 # size of sample 1
|
||||
n2 = 120 # size of sample 2
|
||||
n_resamples = 999 # number of bootstrap resamples used to form each CI
|
||||
confidence_level = 0.9
|
||||
|
||||
# The statistic we're interested in is the difference in means
|
||||
def my_stat(data1, data2, axis=-1):
|
||||
mean1 = np.mean(data1, axis=axis)
|
||||
mean2 = np.mean(data2, axis=axis)
|
||||
return mean1 - mean2
|
||||
|
||||
# The true difference in the means is -0.1
|
||||
dist1 = stats.norm(loc=0, scale=1)
|
||||
dist2 = stats.norm(loc=0.1, scale=1)
|
||||
stat_true = dist1.mean() - dist2.mean()
|
||||
|
||||
# Do the same thing 1000 times. (The code is fully vectorized.)
|
||||
n_replications = 1000
|
||||
data1 = dist1.rvs(size=(n_replications, n1))
|
||||
data2 = dist2.rvs(size=(n_replications, n2))
|
||||
res = bootstrap((data1, data2),
|
||||
statistic=my_stat,
|
||||
confidence_level=confidence_level,
|
||||
n_resamples=n_resamples,
|
||||
batch=50,
|
||||
method=method,
|
||||
axis=-1)
|
||||
ci = res.confidence_interval
|
||||
|
||||
# ci contains vectors of lower and upper confidence interval bounds
|
||||
ci_contains_true = np.sum((ci[0] < stat_true) & (stat_true < ci[1]))
|
||||
assert ci_contains_true == expected
|
||||
|
||||
# ci_contains_true is not inconsistent with confidence_level
|
||||
pvalue = stats.binomtest(ci_contains_true, n_replications,
|
||||
confidence_level).pvalue
|
||||
assert pvalue > 0.1
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ["basic", "percentile"])
|
||||
@pytest.mark.parametrize("axis", [0, 1])
|
||||
def test_bootstrap_vectorized_3samp(method, axis):
|
||||
def statistic(*data, axis=0):
|
||||
# an arbitrary, vectorized statistic
|
||||
return sum((sample.mean(axis) for sample in data))
|
||||
|
||||
def statistic_1d(*data):
|
||||
# the same statistic, not vectorized
|
||||
for sample in data:
|
||||
assert sample.ndim == 1
|
||||
return statistic(*data, axis=0)
|
||||
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(4, 5)
|
||||
y = np.random.rand(4, 5)
|
||||
z = np.random.rand(4, 5)
|
||||
res1 = bootstrap((x, y, z), statistic, vectorized=True,
|
||||
axis=axis, n_resamples=100, method=method, random_state=0)
|
||||
res2 = bootstrap((x, y, z), statistic_1d, vectorized=False,
|
||||
axis=axis, n_resamples=100, method=method, random_state=0)
|
||||
assert_allclose(res1.confidence_interval, res2.confidence_interval)
|
||||
assert_allclose(res1.standard_error, res2.standard_error)
|
||||
|
||||
|
||||
@pytest.mark.xfail_on_32bit("Failure is not concerning; see gh-14107")
|
||||
@pytest.mark.parametrize("method", ["basic", "percentile", "BCa"])
|
||||
@pytest.mark.parametrize("axis", [0, 1])
|
||||
def test_bootstrap_vectorized_1samp(method, axis):
|
||||
def statistic(x, axis=0):
|
||||
# an arbitrary, vectorized statistic
|
||||
return x.mean(axis=axis)
|
||||
|
||||
def statistic_1d(x):
|
||||
# the same statistic, not vectorized
|
||||
assert x.ndim == 1
|
||||
return statistic(x, axis=0)
|
||||
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(4, 5)
|
||||
res1 = bootstrap((x,), statistic, vectorized=True, axis=axis,
|
||||
n_resamples=100, batch=None, method=method,
|
||||
random_state=0)
|
||||
res2 = bootstrap((x,), statistic_1d, vectorized=False, axis=axis,
|
||||
n_resamples=100, batch=10, method=method,
|
||||
random_state=0)
|
||||
assert_allclose(res1.confidence_interval, res2.confidence_interval)
|
||||
assert_allclose(res1.standard_error, res2.standard_error)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ["basic", "percentile", "BCa"])
|
||||
def test_bootstrap_degenerate(method):
|
||||
data = 35 * [10000.]
|
||||
if method == "BCa":
|
||||
with np.errstate(invalid='ignore'):
|
||||
with pytest.warns(BootstrapDegenerateDistributionWarning):
|
||||
res = bootstrap([data, ], np.mean, method=method)
|
||||
assert_equal(res.confidence_interval, (np.nan, np.nan))
|
||||
else:
|
||||
res = bootstrap([data, ], np.mean, method=method)
|
||||
assert_equal(res.confidence_interval, (10000., 10000.))
|
||||
assert_equal(res.standard_error, 0)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("method", ["basic", "percentile", "BCa"])
|
||||
def test_bootstrap_gh15678(method):
|
||||
# Check that gh-15678 is fixed: when statistic function returned a Python
|
||||
# float, method="BCa" failed when trying to add a dimension to the float
|
||||
rng = np.random.default_rng(354645618886684)
|
||||
dist = stats.norm(loc=2, scale=4)
|
||||
data = dist.rvs(size=100, random_state=rng)
|
||||
data = (data,)
|
||||
res = bootstrap(data, stats.skew, method=method, n_resamples=100,
|
||||
random_state=np.random.default_rng(9563))
|
||||
# this always worked because np.apply_along_axis returns NumPy data type
|
||||
ref = bootstrap(data, stats.skew, method=method, n_resamples=100,
|
||||
random_state=np.random.default_rng(9563), vectorized=False)
|
||||
assert_allclose(res.confidence_interval, ref.confidence_interval)
|
||||
assert_allclose(res.standard_error, ref.standard_error)
|
||||
assert isinstance(res.standard_error, np.float64)
|
||||
|
||||
|
||||
def test_jackknife_resample():
|
||||
shape = 3, 4, 5, 6
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(*shape)
|
||||
y = next(_bootstrap._jackknife_resample(x))
|
||||
|
||||
for i in range(shape[-1]):
|
||||
# each resample is indexed along second to last axis
|
||||
# (last axis is the one the statistic will be taken over / consumed)
|
||||
slc = y[..., i, :]
|
||||
expected = np.delete(x, i, axis=-1)
|
||||
|
||||
assert np.array_equal(slc, expected)
|
||||
|
||||
y2 = np.concatenate(list(_bootstrap._jackknife_resample(x, batch=2)),
|
||||
axis=-2)
|
||||
assert np.array_equal(y2, y)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("rng_name", ["RandomState", "default_rng"])
|
||||
def test_bootstrap_resample(rng_name):
|
||||
rng = getattr(np.random, rng_name, None)
|
||||
if rng is None:
|
||||
pytest.skip(f"{rng_name} not available.")
|
||||
rng1 = rng(0)
|
||||
rng2 = rng(0)
|
||||
|
||||
n_resamples = 10
|
||||
shape = 3, 4, 5, 6
|
||||
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(*shape)
|
||||
y = _bootstrap._bootstrap_resample(x, n_resamples, random_state=rng1)
|
||||
|
||||
for i in range(n_resamples):
|
||||
# each resample is indexed along second to last axis
|
||||
# (last axis is the one the statistic will be taken over / consumed)
|
||||
slc = y[..., i, :]
|
||||
|
||||
js = rng_integers(rng2, 0, shape[-1], shape[-1])
|
||||
expected = x[..., js]
|
||||
|
||||
assert np.array_equal(slc, expected)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("score", [0, 0.5, 1])
|
||||
@pytest.mark.parametrize("axis", [0, 1, 2])
|
||||
def test_percentile_of_score(score, axis):
|
||||
shape = 10, 20, 30
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(*shape)
|
||||
p = _bootstrap._percentile_of_score(x, score, axis=-1)
|
||||
|
||||
def vectorized_pos(a, score, axis):
|
||||
return np.apply_along_axis(stats.percentileofscore, axis, a, score)
|
||||
|
||||
p2 = vectorized_pos(x, score, axis=-1)/100
|
||||
|
||||
assert_allclose(p, p2, 1e-15)
|
||||
|
||||
|
||||
def test_percentile_along_axis():
|
||||
# the difference between _percentile_along_axis and np.percentile is that
|
||||
# np.percentile gets _all_ the qs for each axis slice, whereas
|
||||
# _percentile_along_axis gets the q corresponding with each axis slice
|
||||
|
||||
shape = 10, 20
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(*shape)
|
||||
q = np.random.rand(*shape[:-1]) * 100
|
||||
y = _bootstrap._percentile_along_axis(x, q)
|
||||
|
||||
for i in range(shape[0]):
|
||||
res = y[i]
|
||||
expected = np.percentile(x[i], q[i], axis=-1)
|
||||
assert_allclose(res, expected, 1e-15)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("axis", [0, 1, 2])
|
||||
def test_vectorize_statistic(axis):
|
||||
# test that _vectorize_statistic vectorizes a statistic along `axis`
|
||||
|
||||
def statistic(*data, axis):
|
||||
# an arbitrary, vectorized statistic
|
||||
return sum((sample.mean(axis) for sample in data))
|
||||
|
||||
def statistic_1d(*data):
|
||||
# the same statistic, not vectorized
|
||||
for sample in data:
|
||||
assert sample.ndim == 1
|
||||
return statistic(*data, axis=0)
|
||||
|
||||
# vectorize the non-vectorized statistic
|
||||
statistic2 = _bootstrap._vectorize_statistic(statistic_1d)
|
||||
|
||||
np.random.seed(0)
|
||||
x = np.random.rand(4, 5, 6)
|
||||
y = np.random.rand(4, 1, 6)
|
||||
z = np.random.rand(1, 5, 6)
|
||||
|
||||
res1 = statistic(x, y, z, axis=axis)
|
||||
res2 = statistic2(x, y, z, axis=axis)
|
||||
assert_allclose(res1, res2)
|
||||
@@ -0,0 +1,234 @@
|
||||
import numpy as np
|
||||
from numpy.testing import (assert_equal, assert_array_equal,
|
||||
assert_array_almost_equal, assert_approx_equal,
|
||||
assert_allclose)
|
||||
import pytest
|
||||
from pytest import raises as assert_raises
|
||||
from scipy.special import xlogy
|
||||
from scipy.stats.contingency import (margins, expected_freq,
|
||||
chi2_contingency, association)
|
||||
|
||||
|
||||
def test_margins():
|
||||
a = np.array([1])
|
||||
m = margins(a)
|
||||
assert_equal(len(m), 1)
|
||||
m0 = m[0]
|
||||
assert_array_equal(m0, np.array([1]))
|
||||
|
||||
a = np.array([[1]])
|
||||
m0, m1 = margins(a)
|
||||
expected0 = np.array([[1]])
|
||||
expected1 = np.array([[1]])
|
||||
assert_array_equal(m0, expected0)
|
||||
assert_array_equal(m1, expected1)
|
||||
|
||||
a = np.arange(12).reshape(2, 6)
|
||||
m0, m1 = margins(a)
|
||||
expected0 = np.array([[15], [51]])
|
||||
expected1 = np.array([[6, 8, 10, 12, 14, 16]])
|
||||
assert_array_equal(m0, expected0)
|
||||
assert_array_equal(m1, expected1)
|
||||
|
||||
a = np.arange(24).reshape(2, 3, 4)
|
||||
m0, m1, m2 = margins(a)
|
||||
expected0 = np.array([[[66]], [[210]]])
|
||||
expected1 = np.array([[[60], [92], [124]]])
|
||||
expected2 = np.array([[[60, 66, 72, 78]]])
|
||||
assert_array_equal(m0, expected0)
|
||||
assert_array_equal(m1, expected1)
|
||||
assert_array_equal(m2, expected2)
|
||||
|
||||
|
||||
def test_expected_freq():
|
||||
assert_array_equal(expected_freq([1]), np.array([1.0]))
|
||||
|
||||
observed = np.array([[[2, 0], [0, 2]], [[0, 2], [2, 0]], [[1, 1], [1, 1]]])
|
||||
e = expected_freq(observed)
|
||||
assert_array_equal(e, np.ones_like(observed))
|
||||
|
||||
observed = np.array([[10, 10, 20], [20, 20, 20]])
|
||||
e = expected_freq(observed)
|
||||
correct = np.array([[12., 12., 16.], [18., 18., 24.]])
|
||||
assert_array_almost_equal(e, correct)
|
||||
|
||||
|
||||
def test_chi2_contingency_trivial():
|
||||
# Some very simple tests for chi2_contingency.
|
||||
|
||||
# A trivial case
|
||||
obs = np.array([[1, 2], [1, 2]])
|
||||
chi2, p, dof, expected = chi2_contingency(obs, correction=False)
|
||||
assert_equal(chi2, 0.0)
|
||||
assert_equal(p, 1.0)
|
||||
assert_equal(dof, 1)
|
||||
assert_array_equal(obs, expected)
|
||||
|
||||
# A *really* trivial case: 1-D data.
|
||||
obs = np.array([1, 2, 3])
|
||||
chi2, p, dof, expected = chi2_contingency(obs, correction=False)
|
||||
assert_equal(chi2, 0.0)
|
||||
assert_equal(p, 1.0)
|
||||
assert_equal(dof, 0)
|
||||
assert_array_equal(obs, expected)
|
||||
|
||||
|
||||
def test_chi2_contingency_R():
|
||||
# Some test cases that were computed independently, using R.
|
||||
|
||||
# Rcode = \
|
||||
# """
|
||||
# # Data vector.
|
||||
# data <- c(
|
||||
# 12, 34, 23, 4, 47, 11,
|
||||
# 35, 31, 11, 34, 10, 18,
|
||||
# 12, 32, 9, 18, 13, 19,
|
||||
# 12, 12, 14, 9, 33, 25
|
||||
# )
|
||||
#
|
||||
# # Create factor tags:r=rows, c=columns, t=tiers
|
||||
# r <- factor(gl(4, 2*3, 2*3*4, labels=c("r1", "r2", "r3", "r4")))
|
||||
# c <- factor(gl(3, 1, 2*3*4, labels=c("c1", "c2", "c3")))
|
||||
# t <- factor(gl(2, 3, 2*3*4, labels=c("t1", "t2")))
|
||||
#
|
||||
# # 3-way Chi squared test of independence
|
||||
# s = summary(xtabs(data~r+c+t))
|
||||
# print(s)
|
||||
# """
|
||||
# Routput = \
|
||||
# """
|
||||
# Call: xtabs(formula = data ~ r + c + t)
|
||||
# Number of cases in table: 478
|
||||
# Number of factors: 3
|
||||
# Test for independence of all factors:
|
||||
# Chisq = 102.17, df = 17, p-value = 3.514e-14
|
||||
# """
|
||||
obs = np.array(
|
||||
[[[12, 34, 23],
|
||||
[35, 31, 11],
|
||||
[12, 32, 9],
|
||||
[12, 12, 14]],
|
||||
[[4, 47, 11],
|
||||
[34, 10, 18],
|
||||
[18, 13, 19],
|
||||
[9, 33, 25]]])
|
||||
chi2, p, dof, expected = chi2_contingency(obs)
|
||||
assert_approx_equal(chi2, 102.17, significant=5)
|
||||
assert_approx_equal(p, 3.514e-14, significant=4)
|
||||
assert_equal(dof, 17)
|
||||
|
||||
# Rcode = \
|
||||
# """
|
||||
# # Data vector.
|
||||
# data <- c(
|
||||
# #
|
||||
# 12, 17,
|
||||
# 11, 16,
|
||||
# #
|
||||
# 11, 12,
|
||||
# 15, 16,
|
||||
# #
|
||||
# 23, 15,
|
||||
# 30, 22,
|
||||
# #
|
||||
# 14, 17,
|
||||
# 15, 16
|
||||
# )
|
||||
#
|
||||
# # Create factor tags:r=rows, c=columns, d=depths(?), t=tiers
|
||||
# r <- factor(gl(2, 2, 2*2*2*2, labels=c("r1", "r2")))
|
||||
# c <- factor(gl(2, 1, 2*2*2*2, labels=c("c1", "c2")))
|
||||
# d <- factor(gl(2, 4, 2*2*2*2, labels=c("d1", "d2")))
|
||||
# t <- factor(gl(2, 8, 2*2*2*2, labels=c("t1", "t2")))
|
||||
#
|
||||
# # 4-way Chi squared test of independence
|
||||
# s = summary(xtabs(data~r+c+d+t))
|
||||
# print(s)
|
||||
# """
|
||||
# Routput = \
|
||||
# """
|
||||
# Call: xtabs(formula = data ~ r + c + d + t)
|
||||
# Number of cases in table: 262
|
||||
# Number of factors: 4
|
||||
# Test for independence of all factors:
|
||||
# Chisq = 8.758, df = 11, p-value = 0.6442
|
||||
# """
|
||||
obs = np.array(
|
||||
[[[[12, 17],
|
||||
[11, 16]],
|
||||
[[11, 12],
|
||||
[15, 16]]],
|
||||
[[[23, 15],
|
||||
[30, 22]],
|
||||
[[14, 17],
|
||||
[15, 16]]]])
|
||||
chi2, p, dof, expected = chi2_contingency(obs)
|
||||
assert_approx_equal(chi2, 8.758, significant=4)
|
||||
assert_approx_equal(p, 0.6442, significant=4)
|
||||
assert_equal(dof, 11)
|
||||
|
||||
|
||||
def test_chi2_contingency_g():
|
||||
c = np.array([[15, 60], [15, 90]])
|
||||
g, p, dof, e = chi2_contingency(c, lambda_='log-likelihood',
|
||||
correction=False)
|
||||
assert_allclose(g, 2*xlogy(c, c/e).sum())
|
||||
|
||||
g, p, dof, e = chi2_contingency(c, lambda_='log-likelihood',
|
||||
correction=True)
|
||||
c_corr = c + np.array([[-0.5, 0.5], [0.5, -0.5]])
|
||||
assert_allclose(g, 2*xlogy(c_corr, c_corr/e).sum())
|
||||
|
||||
c = np.array([[10, 12, 10], [12, 10, 10]])
|
||||
g, p, dof, e = chi2_contingency(c, lambda_='log-likelihood')
|
||||
assert_allclose(g, 2*xlogy(c, c/e).sum())
|
||||
|
||||
|
||||
def test_chi2_contingency_bad_args():
|
||||
# Test that "bad" inputs raise a ValueError.
|
||||
|
||||
# Negative value in the array of observed frequencies.
|
||||
obs = np.array([[-1, 10], [1, 2]])
|
||||
assert_raises(ValueError, chi2_contingency, obs)
|
||||
|
||||
# The zeros in this will result in zeros in the array
|
||||
# of expected frequencies.
|
||||
obs = np.array([[0, 1], [0, 1]])
|
||||
assert_raises(ValueError, chi2_contingency, obs)
|
||||
|
||||
# A degenerate case: `observed` has size 0.
|
||||
obs = np.empty((0, 8))
|
||||
assert_raises(ValueError, chi2_contingency, obs)
|
||||
|
||||
|
||||
def test_chi2_contingency_yates_gh13875():
|
||||
# Magnitude of Yates' continuity correction should not exceed difference
|
||||
# between expected and observed value of the statistic; see gh-13875
|
||||
observed = np.array([[1573, 3], [4, 0]])
|
||||
p = chi2_contingency(observed)[1]
|
||||
assert_allclose(p, 1, rtol=1e-12)
|
||||
|
||||
|
||||
def test_bad_association_args():
|
||||
# Invalid Test Statistic
|
||||
assert_raises(ValueError, association, [[1, 2], [3, 4]], "X")
|
||||
# Invalid array shape
|
||||
assert_raises(ValueError, association, [[[1, 2]], [[3, 4]]], "cramer")
|
||||
# chi2_contingency exception
|
||||
assert_raises(ValueError, association, [[-1, 10], [1, 2]], 'cramer')
|
||||
# Invalid Array Item Data Type
|
||||
assert_raises(ValueError, association,
|
||||
np.array([[1, 2], ["dd", 4]], dtype=object), 'cramer')
|
||||
|
||||
|
||||
@pytest.mark.parametrize('stat, expected',
|
||||
[('cramer', 0.09222412010290792),
|
||||
('tschuprow', 0.0775509319944633),
|
||||
('pearson', 0.12932925727138758)])
|
||||
def test_assoc(stat, expected):
|
||||
# 2d Array
|
||||
obs1 = np.array([[12, 13, 14, 15, 16],
|
||||
[17, 16, 18, 19, 11],
|
||||
[9, 15, 14, 12, 11]])
|
||||
a = association(observed=obs1, method=stat)
|
||||
assert_allclose(a, expected)
|
||||
@@ -0,0 +1,834 @@
|
||||
|
||||
import numpy as np
|
||||
import numpy.testing as npt
|
||||
import pytest
|
||||
from pytest import raises as assert_raises
|
||||
from scipy.integrate import IntegrationWarning
|
||||
|
||||
from scipy import stats
|
||||
from scipy.special import betainc
|
||||
from .common_tests import (check_normalization, check_moment, check_mean_expect,
|
||||
check_var_expect, check_skew_expect,
|
||||
check_kurt_expect, check_entropy,
|
||||
check_private_entropy, check_entropy_vect_scale,
|
||||
check_edge_support, check_named_args,
|
||||
check_random_state_property,
|
||||
check_meth_dtype, check_ppf_dtype, check_cmplx_deriv,
|
||||
check_pickling, check_rvs_broadcast, check_freezing)
|
||||
from scipy.stats._distr_params import distcont
|
||||
|
||||
"""
|
||||
Test all continuous distributions.
|
||||
|
||||
Parameters were chosen for those distributions that pass the
|
||||
Kolmogorov-Smirnov test. This provides safe parameters for each
|
||||
distributions so that we can perform further testing of class methods.
|
||||
|
||||
These tests currently check only/mostly for serious errors and exceptions,
|
||||
not for numerically exact results.
|
||||
"""
|
||||
|
||||
# Note that you need to add new distributions you want tested
|
||||
# to _distr_params
|
||||
|
||||
DECIMAL = 5 # specify the precision of the tests # increased from 0 to 5
|
||||
|
||||
distslow = ['kstwo', 'genexpon', 'ksone', 'recipinvgauss', 'vonmises',
|
||||
'kappa4', 'vonmises_line', 'gausshyper', 'norminvgauss',
|
||||
'geninvgauss', 'genhyperbolic']
|
||||
# distslow are sorted by speed (very slow to slow)
|
||||
|
||||
distxslow = ['studentized_range']
|
||||
# distxslow are sorted by speed (very slow to slow)
|
||||
|
||||
# skip check_fit_args (test is slow)
|
||||
skip_fit_test_mle = ['exponpow', 'exponweib', 'gausshyper', 'genexpon',
|
||||
'halfgennorm', 'gompertz', 'johnsonsb', 'johnsonsu',
|
||||
'kappa4', 'ksone', 'kstwo', 'kstwobign', 'mielke', 'ncf',
|
||||
'nct', 'powerlognorm', 'powernorm', 'recipinvgauss',
|
||||
'trapezoid', 'vonmises', 'vonmises_line', 'levy_stable',
|
||||
'rv_histogram_instance', 'studentized_range']
|
||||
|
||||
# these were really slow in `test_fit`.py.
|
||||
# note that this list is used to skip both fit_test and fit_fix tests
|
||||
slow_fit_test_mm = ['argus', 'exponpow', 'exponweib', 'gausshyper', 'genexpon',
|
||||
'genhalflogistic', 'halfgennorm', 'gompertz', 'johnsonsb',
|
||||
'kappa4', 'kstwobign', 'recipinvgauss', 'skewnorm',
|
||||
'trapezoid', 'truncexpon', 'vonmises', 'vonmises_line',
|
||||
'studentized_range']
|
||||
# pearson3 fails due to something weird
|
||||
# the first list fails due to non-finite distribution moments encountered
|
||||
# most of the rest fail due to integration warnings
|
||||
# pearson3 is overriden as not implemented due to gh-11746
|
||||
fail_fit_test_mm = (['alpha', 'betaprime', 'bradford', 'burr', 'burr12',
|
||||
'cauchy', 'crystalball', 'f', 'fisk', 'foldcauchy',
|
||||
'genextreme', 'genpareto', 'halfcauchy', 'invgamma',
|
||||
'kappa3', 'levy', 'levy_l', 'loglaplace', 'lomax',
|
||||
'mielke', 'nakagami', 'ncf', 'skewcauchy', 't',
|
||||
'tukeylambda', 'invweibull']
|
||||
+ ['genhyperbolic', 'johnsonsu', 'ksone', 'kstwo',
|
||||
'nct', 'pareto', 'powernorm', 'powerlognorm']
|
||||
+ ['pearson3'])
|
||||
skip_fit_test = {"MLE": skip_fit_test_mle,
|
||||
"MM": slow_fit_test_mm + fail_fit_test_mm}
|
||||
|
||||
# skip check_fit_args_fix (test is slow)
|
||||
skip_fit_fix_test_mle = ['burr', 'exponpow', 'exponweib', 'gausshyper',
|
||||
'genexpon', 'halfgennorm', 'gompertz', 'johnsonsb',
|
||||
'johnsonsu', 'kappa4', 'ksone', 'kstwo', 'kstwobign',
|
||||
'levy_stable', 'mielke', 'ncf', 'ncx2',
|
||||
'powerlognorm', 'powernorm', 'rdist', 'recipinvgauss',
|
||||
'trapezoid', 'vonmises', 'vonmises_line',
|
||||
'studentized_range']
|
||||
# the first list fails due to non-finite distribution moments encountered
|
||||
# most of the rest fail due to integration warnings
|
||||
# pearson3 is overriden as not implemented due to gh-11746
|
||||
fail_fit_fix_test_mm = (['alpha', 'betaprime', 'burr', 'burr12', 'cauchy',
|
||||
'crystalball', 'f', 'fisk', 'foldcauchy',
|
||||
'genextreme', 'genpareto', 'halfcauchy', 'invgamma',
|
||||
'kappa3', 'levy', 'levy_l', 'loglaplace', 'lomax',
|
||||
'mielke', 'nakagami', 'ncf', 'nct', 'skewcauchy', 't',
|
||||
'invweibull']
|
||||
+ ['genhyperbolic', 'johnsonsu', 'ksone', 'kstwo',
|
||||
'pareto', 'powernorm', 'powerlognorm']
|
||||
+ ['pearson3'])
|
||||
skip_fit_fix_test = {"MLE": skip_fit_fix_test_mle,
|
||||
"MM": slow_fit_test_mm + fail_fit_fix_test_mm}
|
||||
|
||||
# These distributions fail the complex derivative test below.
|
||||
# Here 'fail' mean produce wrong results and/or raise exceptions, depending
|
||||
# on the implementation details of corresponding special functions.
|
||||
# cf https://github.com/scipy/scipy/pull/4979 for a discussion.
|
||||
fails_cmplx = set(['argus', 'beta', 'betaprime', 'chi', 'chi2', 'cosine',
|
||||
'dgamma', 'dweibull', 'erlang', 'f', 'gamma',
|
||||
'gausshyper', 'gengamma', 'genhyperbolic',
|
||||
'geninvgauss', 'gennorm', 'genpareto',
|
||||
'halfgennorm', 'invgamma',
|
||||
'ksone', 'kstwo', 'kstwobign', 'levy_l', 'loggamma',
|
||||
'logistic', 'loguniform', 'maxwell', 'nakagami',
|
||||
'ncf', 'nct', 'ncx2', 'norminvgauss', 'pearson3', 'rdist',
|
||||
'reciprocal', 'rice', 'skewnorm', 't', 'tukeylambda',
|
||||
'vonmises', 'vonmises_line', 'rv_histogram_instance',
|
||||
'truncnorm', 'studentized_range'])
|
||||
|
||||
_h = np.histogram([1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 5, 6,
|
||||
6, 6, 6, 7, 7, 7, 8, 8, 9], bins=8)
|
||||
histogram_test_instance = stats.rv_histogram(_h)
|
||||
|
||||
|
||||
def cases_test_cont_basic():
|
||||
for distname, arg in distcont[:] + [(histogram_test_instance, tuple())]:
|
||||
if distname == 'levy_stable':
|
||||
continue
|
||||
elif distname in distslow:
|
||||
yield pytest.param(distname, arg, marks=pytest.mark.slow)
|
||||
elif distname in distxslow:
|
||||
yield pytest.param(distname, arg, marks=pytest.mark.xslow)
|
||||
else:
|
||||
yield distname, arg
|
||||
|
||||
|
||||
@pytest.mark.parametrize('distname,arg', cases_test_cont_basic())
|
||||
@pytest.mark.parametrize('sn, n_fit_samples', [(500, 200)])
|
||||
def test_cont_basic(distname, arg, sn, n_fit_samples):
|
||||
# this test skips slow distributions
|
||||
|
||||
try:
|
||||
distfn = getattr(stats, distname)
|
||||
except TypeError:
|
||||
distfn = distname
|
||||
distname = 'rv_histogram_instance'
|
||||
|
||||
rng = np.random.RandomState(765456)
|
||||
rvs = distfn.rvs(size=sn, *arg, random_state=rng)
|
||||
sm = rvs.mean()
|
||||
sv = rvs.var()
|
||||
m, v = distfn.stats(*arg)
|
||||
|
||||
check_sample_meanvar_(distfn, arg, m, v, sm, sv, sn, distname + 'sample mean test')
|
||||
check_cdf_ppf(distfn, arg, distname)
|
||||
check_sf_isf(distfn, arg, distname)
|
||||
check_pdf(distfn, arg, distname)
|
||||
check_pdf_logpdf(distfn, arg, distname)
|
||||
check_pdf_logpdf_at_endpoints(distfn, arg, distname)
|
||||
check_cdf_logcdf(distfn, arg, distname)
|
||||
check_sf_logsf(distfn, arg, distname)
|
||||
check_ppf_broadcast(distfn, arg, distname)
|
||||
|
||||
alpha = 0.01
|
||||
if distname == 'rv_histogram_instance':
|
||||
check_distribution_rvs(distfn.cdf, arg, alpha, rvs)
|
||||
elif distname != 'geninvgauss':
|
||||
# skip kstest for geninvgauss since cdf is too slow; see test for
|
||||
# rv generation in TestGenInvGauss in test_distributions.py
|
||||
check_distribution_rvs(distname, arg, alpha, rvs)
|
||||
|
||||
locscale_defaults = (0, 1)
|
||||
meths = [distfn.pdf, distfn.logpdf, distfn.cdf, distfn.logcdf,
|
||||
distfn.logsf]
|
||||
# make sure arguments are within support
|
||||
spec_x = {'weibull_max': -0.5, 'levy_l': -0.5,
|
||||
'pareto': 1.5, 'tukeylambda': 0.3,
|
||||
'rv_histogram_instance': 5.0}
|
||||
x = spec_x.get(distname, 0.5)
|
||||
if distname == 'invweibull':
|
||||
arg = (1,)
|
||||
elif distname == 'ksone':
|
||||
arg = (3,)
|
||||
|
||||
check_named_args(distfn, x, arg, locscale_defaults, meths)
|
||||
check_random_state_property(distfn, arg)
|
||||
check_pickling(distfn, arg)
|
||||
check_freezing(distfn, arg)
|
||||
|
||||
# Entropy
|
||||
if distname not in ['kstwobign', 'kstwo']:
|
||||
check_entropy(distfn, arg, distname)
|
||||
|
||||
if distfn.numargs == 0:
|
||||
check_vecentropy(distfn, arg)
|
||||
|
||||
if (distfn.__class__._entropy != stats.rv_continuous._entropy
|
||||
and distname != 'vonmises'):
|
||||
check_private_entropy(distfn, arg, stats.rv_continuous)
|
||||
|
||||
with npt.suppress_warnings() as sup:
|
||||
sup.filter(IntegrationWarning, "The occurrence of roundoff error")
|
||||
sup.filter(IntegrationWarning, "Extremely bad integrand")
|
||||
sup.filter(RuntimeWarning, "invalid value")
|
||||
check_entropy_vect_scale(distfn, arg)
|
||||
|
||||
check_retrieving_support(distfn, arg)
|
||||
check_edge_support(distfn, arg)
|
||||
|
||||
check_meth_dtype(distfn, arg, meths)
|
||||
check_ppf_dtype(distfn, arg)
|
||||
|
||||
if distname not in fails_cmplx:
|
||||
check_cmplx_deriv(distfn, arg)
|
||||
|
||||
if distname != 'truncnorm':
|
||||
check_ppf_private(distfn, arg, distname)
|
||||
|
||||
for method in ["MLE", "MM"]:
|
||||
if distname not in skip_fit_test[method]:
|
||||
check_fit_args(distfn, arg, rvs[:n_fit_samples], method)
|
||||
|
||||
if distname not in skip_fit_fix_test[method]:
|
||||
check_fit_args_fix(distfn, arg, rvs[:n_fit_samples], method)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('distname,arg', cases_test_cont_basic())
|
||||
def test_rvs_scalar(distname, arg):
|
||||
# rvs should return a scalar when given scalar arguments (gh-12428)
|
||||
try:
|
||||
distfn = getattr(stats, distname)
|
||||
except TypeError:
|
||||
distfn = distname
|
||||
distname = 'rv_histogram_instance'
|
||||
|
||||
assert np.isscalar(distfn.rvs(*arg))
|
||||
assert np.isscalar(distfn.rvs(*arg, size=()))
|
||||
assert np.isscalar(distfn.rvs(*arg, size=None))
|
||||
|
||||
|
||||
def test_levy_stable_random_state_property():
|
||||
# levy_stable only implements rvs(), so it is skipped in the
|
||||
# main loop in test_cont_basic(). Here we apply just the test
|
||||
# check_random_state_property to levy_stable.
|
||||
check_random_state_property(stats.levy_stable, (0.5, 0.1))
|
||||
|
||||
|
||||
def cases_test_moments():
|
||||
fail_normalization = set(['vonmises'])
|
||||
fail_higher = set(['vonmises', 'ncf'])
|
||||
fail_loc_scale = set(['kappa3', 'kappa4']) # see gh-13582
|
||||
|
||||
for distname, arg in distcont[:] + [(histogram_test_instance, tuple())]:
|
||||
if distname == 'levy_stable':
|
||||
continue
|
||||
|
||||
if distname == 'studentized_range':
|
||||
msg = ("studentized_range is far too slow for this test and it is "
|
||||
"redundant with test_distributions::TestStudentizedRange::"
|
||||
"test_moment_against_mp")
|
||||
yield pytest.param(distname, arg, True, True, True, True,
|
||||
marks=pytest.mark.xslow(reason=msg))
|
||||
continue
|
||||
cond1 = distname not in fail_normalization
|
||||
cond2 = distname not in fail_higher
|
||||
cond3 = distname not in fail_loc_scale
|
||||
|
||||
yield distname, arg, cond1, cond2, cond3, False
|
||||
|
||||
if not cond1 or not cond2 or not cond3:
|
||||
# Run the distributions that have issues twice, once skipping the
|
||||
# not_ok parts, once with the not_ok parts but marked as knownfail
|
||||
yield pytest.param(distname, arg, True, True, True, True,
|
||||
marks=pytest.mark.xfail)
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
@pytest.mark.parametrize('distname,arg,normalization_ok,higher_ok,'
|
||||
'loc_scale_ok,is_xfailing',
|
||||
cases_test_moments())
|
||||
def test_moments(distname, arg, normalization_ok, higher_ok, loc_scale_ok,
|
||||
is_xfailing):
|
||||
try:
|
||||
distfn = getattr(stats, distname)
|
||||
except TypeError:
|
||||
distfn = distname
|
||||
distname = 'rv_histogram_instance'
|
||||
|
||||
with npt.suppress_warnings() as sup:
|
||||
sup.filter(IntegrationWarning,
|
||||
"The integral is probably divergent, or slowly convergent.")
|
||||
if is_xfailing:
|
||||
sup.filter(IntegrationWarning)
|
||||
|
||||
m, v, s, k = distfn.stats(*arg, moments='mvsk')
|
||||
|
||||
if normalization_ok:
|
||||
check_normalization(distfn, arg, distname)
|
||||
|
||||
if higher_ok:
|
||||
check_mean_expect(distfn, arg, m, distname)
|
||||
check_skew_expect(distfn, arg, m, v, s, distname)
|
||||
check_var_expect(distfn, arg, m, v, distname)
|
||||
check_kurt_expect(distfn, arg, m, v, k, distname)
|
||||
|
||||
if loc_scale_ok:
|
||||
check_loc_scale(distfn, arg, m, v, distname)
|
||||
|
||||
check_moment(distfn, arg, m, v, distname)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('dist,shape_args', distcont)
|
||||
def test_rvs_broadcast(dist, shape_args):
|
||||
if dist in ['gausshyper', 'genexpon', 'studentized_range']:
|
||||
pytest.skip("too slow")
|
||||
|
||||
# If shape_only is True, it means the _rvs method of the
|
||||
# distribution uses more than one random number to generate a random
|
||||
# variate. That means the result of using rvs with broadcasting or
|
||||
# with a nontrivial size will not necessarily be the same as using the
|
||||
# numpy.vectorize'd version of rvs(), so we can only compare the shapes
|
||||
# of the results, not the values.
|
||||
# Whether or not a distribution is in the following list is an
|
||||
# implementation detail of the distribution, not a requirement. If
|
||||
# the implementation the rvs() method of a distribution changes, this
|
||||
# test might also have to be changed.
|
||||
shape_only = dist in ['argus', 'betaprime', 'dgamma', 'dweibull',
|
||||
'exponnorm', 'genhyperbolic', 'geninvgauss',
|
||||
'levy_stable', 'nct', 'norminvgauss', 'rice',
|
||||
'skewnorm', 'semicircular']
|
||||
|
||||
distfunc = getattr(stats, dist)
|
||||
loc = np.zeros(2)
|
||||
scale = np.ones((3, 1))
|
||||
nargs = distfunc.numargs
|
||||
allargs = []
|
||||
bshape = [3, 2]
|
||||
# Generate shape parameter arguments...
|
||||
for k in range(nargs):
|
||||
shp = (k + 4,) + (1,)*(k + 2)
|
||||
allargs.append(shape_args[k]*np.ones(shp))
|
||||
bshape.insert(0, k + 4)
|
||||
allargs.extend([loc, scale])
|
||||
# bshape holds the expected shape when loc, scale, and the shape
|
||||
# parameters are all broadcast together.
|
||||
|
||||
check_rvs_broadcast(distfunc, dist, allargs, bshape, shape_only, 'd')
|
||||
|
||||
|
||||
def test_rvs_gh2069_regression():
|
||||
# Regression tests for gh-2069. In scipy 0.17 and earlier,
|
||||
# these tests would fail.
|
||||
#
|
||||
# A typical example of the broken behavior:
|
||||
# >>> norm.rvs(loc=np.zeros(5), scale=np.ones(5))
|
||||
# array([-2.49613705, -2.49613705, -2.49613705, -2.49613705, -2.49613705])
|
||||
rng = np.random.RandomState(123)
|
||||
vals = stats.norm.rvs(loc=np.zeros(5), scale=1, random_state=rng)
|
||||
d = np.diff(vals)
|
||||
npt.assert_(np.all(d != 0), "All the values are equal, but they shouldn't be!")
|
||||
vals = stats.norm.rvs(loc=0, scale=np.ones(5), random_state=rng)
|
||||
d = np.diff(vals)
|
||||
npt.assert_(np.all(d != 0), "All the values are equal, but they shouldn't be!")
|
||||
vals = stats.norm.rvs(loc=np.zeros(5), scale=np.ones(5), random_state=rng)
|
||||
d = np.diff(vals)
|
||||
npt.assert_(np.all(d != 0), "All the values are equal, but they shouldn't be!")
|
||||
vals = stats.norm.rvs(loc=np.array([[0], [0]]), scale=np.ones(5),
|
||||
random_state=rng)
|
||||
d = np.diff(vals.ravel())
|
||||
npt.assert_(np.all(d != 0), "All the values are equal, but they shouldn't be!")
|
||||
|
||||
assert_raises(ValueError, stats.norm.rvs, [[0, 0], [0, 0]],
|
||||
[[1, 1], [1, 1]], 1)
|
||||
assert_raises(ValueError, stats.gamma.rvs, [2, 3, 4, 5], 0, 1, (2, 2))
|
||||
assert_raises(ValueError, stats.gamma.rvs, [1, 1, 1, 1], [0, 0, 0, 0],
|
||||
[[1], [2]], (4,))
|
||||
|
||||
|
||||
def test_nomodify_gh9900_regression():
|
||||
# Regression test for gh-9990
|
||||
# Prior to gh-9990, calls to stats.truncnorm._cdf() use what ever was
|
||||
# set inside the stats.truncnorm instance during stats.truncnorm.cdf().
|
||||
# This could cause issues wth multi-threaded code.
|
||||
# Since then, the calls to cdf() are not permitted to modify the global
|
||||
# stats.truncnorm instance.
|
||||
tn = stats.truncnorm
|
||||
# Use the right-half truncated normal
|
||||
# Check that the cdf and _cdf return the same result.
|
||||
npt.assert_almost_equal(tn.cdf(1, 0, np.inf), 0.6826894921370859)
|
||||
npt.assert_almost_equal(tn._cdf(1, 0, np.inf), 0.6826894921370859)
|
||||
|
||||
# Now use the left-half truncated normal
|
||||
npt.assert_almost_equal(tn.cdf(-1, -np.inf, 0), 0.31731050786291415)
|
||||
npt.assert_almost_equal(tn._cdf(-1, -np.inf, 0), 0.31731050786291415)
|
||||
|
||||
# Check that the right-half truncated normal _cdf hasn't changed
|
||||
npt.assert_almost_equal(tn._cdf(1, 0, np.inf), 0.6826894921370859) # NOT 1.6826894921370859
|
||||
npt.assert_almost_equal(tn.cdf(1, 0, np.inf), 0.6826894921370859)
|
||||
|
||||
# Check that the left-half truncated normal _cdf hasn't changed
|
||||
npt.assert_almost_equal(tn._cdf(-1, -np.inf, 0), 0.31731050786291415) # Not -0.6826894921370859
|
||||
npt.assert_almost_equal(tn.cdf(1, -np.inf, 0), 1) # Not 1.6826894921370859
|
||||
npt.assert_almost_equal(tn.cdf(-1, -np.inf, 0), 0.31731050786291415) # Not -0.6826894921370859
|
||||
|
||||
|
||||
def test_broadcast_gh9990_regression():
|
||||
# Regression test for gh-9990
|
||||
# The x-value 7 only lies within the support of 4 of the supplied
|
||||
# distributions. Prior to 9990, one array passed to
|
||||
# stats.reciprocal._cdf would have 4 elements, but an array
|
||||
# previously stored by stats.reciprocal_argcheck() would have 6, leading
|
||||
# to a broadcast error.
|
||||
a = np.array([1, 2, 3, 4, 5, 6])
|
||||
b = np.array([8, 16, 1, 32, 1, 48])
|
||||
ans = [stats.reciprocal.cdf(7, _a, _b) for _a, _b in zip(a,b)]
|
||||
npt.assert_array_almost_equal(stats.reciprocal.cdf(7, a, b), ans)
|
||||
|
||||
ans = [stats.reciprocal.cdf(1, _a, _b) for _a, _b in zip(a,b)]
|
||||
npt.assert_array_almost_equal(stats.reciprocal.cdf(1, a, b), ans)
|
||||
|
||||
ans = [stats.reciprocal.cdf(_a, _a, _b) for _a, _b in zip(a,b)]
|
||||
npt.assert_array_almost_equal(stats.reciprocal.cdf(a, a, b), ans)
|
||||
|
||||
ans = [stats.reciprocal.cdf(_b, _a, _b) for _a, _b in zip(a,b)]
|
||||
npt.assert_array_almost_equal(stats.reciprocal.cdf(b, a, b), ans)
|
||||
|
||||
|
||||
def test_broadcast_gh7933_regression():
|
||||
# Check broadcast works
|
||||
stats.truncnorm.logpdf(
|
||||
np.array([3.0, 2.0, 1.0]),
|
||||
a=(1.5 - np.array([6.0, 5.0, 4.0])) / 3.0,
|
||||
b=np.inf,
|
||||
loc=np.array([6.0, 5.0, 4.0]),
|
||||
scale=3.0
|
||||
)
|
||||
|
||||
|
||||
def test_gh2002_regression():
|
||||
# Add a check that broadcast works in situations where only some
|
||||
# x-values are compatible with some of the shape arguments.
|
||||
x = np.r_[-2:2:101j]
|
||||
a = np.r_[-np.ones(50), np.ones(51)]
|
||||
expected = [stats.truncnorm.pdf(_x, _a, np.inf) for _x, _a in zip(x, a)]
|
||||
ans = stats.truncnorm.pdf(x, a, np.inf)
|
||||
npt.assert_array_almost_equal(ans, expected)
|
||||
|
||||
|
||||
def test_gh1320_regression():
|
||||
# Check that the first example from gh-1320 now works.
|
||||
c = 2.62
|
||||
stats.genextreme.ppf(0.5, np.array([[c], [c + 0.5]]))
|
||||
# The other examples in gh-1320 appear to have stopped working
|
||||
# some time ago.
|
||||
# ans = stats.genextreme.moment(2, np.array([c, c + 0.5]))
|
||||
# expected = np.array([25.50105963, 115.11191437])
|
||||
# stats.genextreme.moment(5, np.array([[c], [c + 0.5]]))
|
||||
# stats.genextreme.moment(5, np.array([c, c + 0.5]))
|
||||
|
||||
|
||||
def test_method_of_moments():
|
||||
# example from https://en.wikipedia.org/wiki/Method_of_moments_(statistics)
|
||||
np.random.seed(1234)
|
||||
x = [0, 0, 0, 0, 1]
|
||||
a = 1/5 - 2*np.sqrt(3)/5
|
||||
b = 1/5 + 2*np.sqrt(3)/5
|
||||
# force use of method of moments (uniform.fit is overriden)
|
||||
loc, scale = super(type(stats.uniform), stats.uniform).fit(x, method="MM")
|
||||
npt.assert_almost_equal(loc, a, decimal=4)
|
||||
npt.assert_almost_equal(loc+scale, b, decimal=4)
|
||||
|
||||
|
||||
def check_sample_meanvar_(distfn, arg, m, v, sm, sv, sn, msg):
|
||||
# this did not work, skipped silently by nose
|
||||
if np.isfinite(m):
|
||||
check_sample_mean(sm, sv, sn, m)
|
||||
if np.isfinite(v):
|
||||
check_sample_var(sv, sn, v)
|
||||
|
||||
|
||||
def check_sample_mean(sm, v, n, popmean):
|
||||
# from stats._stats_py.ttest_1samp(a, popmean):
|
||||
# Calculates the t-obtained for the independent samples T-test on ONE group
|
||||
# of scores a, given a population mean.
|
||||
#
|
||||
# Returns: t-value, two-tailed prob
|
||||
df = n-1
|
||||
svar = ((n-1)*v) / float(df) # looks redundant
|
||||
t = (sm-popmean) / np.sqrt(svar*(1.0/n))
|
||||
prob = betainc(0.5*df, 0.5, df/(df + t*t))
|
||||
|
||||
# return t,prob
|
||||
npt.assert_(prob > 0.01, 'mean fail, t,prob = %f, %f, m, sm=%f,%f' %
|
||||
(t, prob, popmean, sm))
|
||||
|
||||
|
||||
def check_sample_var(sv, n, popvar):
|
||||
# two-sided chisquare test for sample variance equal to
|
||||
# hypothesized variance
|
||||
df = n-1
|
||||
chi2 = (n - 1)*sv/popvar
|
||||
pval = stats.distributions.chi2.sf(chi2, df) * 2
|
||||
npt.assert_(pval > 0.01, 'var fail, t, pval = %f, %f, v, sv=%f, %f' %
|
||||
(chi2, pval, popvar, sv))
|
||||
|
||||
|
||||
def check_cdf_ppf(distfn, arg, msg):
|
||||
values = [0.001, 0.5, 0.999]
|
||||
npt.assert_almost_equal(distfn.cdf(distfn.ppf(values, *arg), *arg),
|
||||
values, decimal=DECIMAL, err_msg=msg +
|
||||
' - cdf-ppf roundtrip')
|
||||
|
||||
|
||||
def check_sf_isf(distfn, arg, msg):
|
||||
npt.assert_almost_equal(distfn.sf(distfn.isf([0.1, 0.5, 0.9], *arg), *arg),
|
||||
[0.1, 0.5, 0.9], decimal=DECIMAL, err_msg=msg +
|
||||
' - sf-isf roundtrip')
|
||||
npt.assert_almost_equal(distfn.cdf([0.1, 0.9], *arg),
|
||||
1.0 - distfn.sf([0.1, 0.9], *arg),
|
||||
decimal=DECIMAL, err_msg=msg +
|
||||
' - cdf-sf relationship')
|
||||
|
||||
|
||||
def check_pdf(distfn, arg, msg):
|
||||
# compares pdf at median with numerical derivative of cdf
|
||||
median = distfn.ppf(0.5, *arg)
|
||||
eps = 1e-6
|
||||
pdfv = distfn.pdf(median, *arg)
|
||||
if (pdfv < 1e-4) or (pdfv > 1e4):
|
||||
# avoid checking a case where pdf is close to zero or
|
||||
# huge (singularity)
|
||||
median = median + 0.1
|
||||
pdfv = distfn.pdf(median, *arg)
|
||||
cdfdiff = (distfn.cdf(median + eps, *arg) -
|
||||
distfn.cdf(median - eps, *arg))/eps/2.0
|
||||
# replace with better diff and better test (more points),
|
||||
# actually, this works pretty well
|
||||
msg += ' - cdf-pdf relationship'
|
||||
npt.assert_almost_equal(pdfv, cdfdiff, decimal=DECIMAL, err_msg=msg)
|
||||
|
||||
|
||||
def check_pdf_logpdf(distfn, args, msg):
|
||||
# compares pdf at several points with the log of the pdf
|
||||
points = np.array([0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8])
|
||||
vals = distfn.ppf(points, *args)
|
||||
vals = vals[np.isfinite(vals)]
|
||||
pdf = distfn.pdf(vals, *args)
|
||||
logpdf = distfn.logpdf(vals, *args)
|
||||
pdf = pdf[(pdf != 0) & np.isfinite(pdf)]
|
||||
logpdf = logpdf[np.isfinite(logpdf)]
|
||||
msg += " - logpdf-log(pdf) relationship"
|
||||
npt.assert_almost_equal(np.log(pdf), logpdf, decimal=7, err_msg=msg)
|
||||
|
||||
|
||||
def check_pdf_logpdf_at_endpoints(distfn, args, msg):
|
||||
# compares pdf with the log of the pdf at the (finite) end points
|
||||
points = np.array([0, 1])
|
||||
vals = distfn.ppf(points, *args)
|
||||
vals = vals[np.isfinite(vals)]
|
||||
with npt.suppress_warnings() as sup:
|
||||
# Several distributions incur divide by zero or encounter invalid values when computing
|
||||
# the pdf or logpdf at the endpoints.
|
||||
suppress_messsages = [
|
||||
"divide by zero encountered in true_divide", # multiple distributions
|
||||
"divide by zero encountered in log", # multiple distributions
|
||||
"divide by zero encountered in power", # gengamma
|
||||
"invalid value encountered in add", # genextreme
|
||||
"invalid value encountered in subtract", # gengamma
|
||||
]
|
||||
for msg in suppress_messsages:
|
||||
sup.filter(category=RuntimeWarning, message=msg)
|
||||
|
||||
pdf = distfn.pdf(vals, *args)
|
||||
logpdf = distfn.logpdf(vals, *args)
|
||||
pdf = pdf[(pdf != 0) & np.isfinite(pdf)]
|
||||
logpdf = logpdf[np.isfinite(logpdf)]
|
||||
msg += " - logpdf-log(pdf) relationship"
|
||||
npt.assert_almost_equal(np.log(pdf), logpdf, decimal=7, err_msg=msg)
|
||||
|
||||
|
||||
def check_sf_logsf(distfn, args, msg):
|
||||
# compares sf at several points with the log of the sf
|
||||
points = np.array([0.0, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 1.0])
|
||||
vals = distfn.ppf(points, *args)
|
||||
vals = vals[np.isfinite(vals)]
|
||||
sf = distfn.sf(vals, *args)
|
||||
logsf = distfn.logsf(vals, *args)
|
||||
sf = sf[sf != 0]
|
||||
logsf = logsf[np.isfinite(logsf)]
|
||||
msg += " - logsf-log(sf) relationship"
|
||||
npt.assert_almost_equal(np.log(sf), logsf, decimal=7, err_msg=msg)
|
||||
|
||||
|
||||
def check_cdf_logcdf(distfn, args, msg):
|
||||
# compares cdf at several points with the log of the cdf
|
||||
points = np.array([0, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 1.0])
|
||||
vals = distfn.ppf(points, *args)
|
||||
vals = vals[np.isfinite(vals)]
|
||||
cdf = distfn.cdf(vals, *args)
|
||||
logcdf = distfn.logcdf(vals, *args)
|
||||
cdf = cdf[cdf != 0]
|
||||
logcdf = logcdf[np.isfinite(logcdf)]
|
||||
msg += " - logcdf-log(cdf) relationship"
|
||||
npt.assert_almost_equal(np.log(cdf), logcdf, decimal=7, err_msg=msg)
|
||||
|
||||
|
||||
def check_ppf_broadcast(distfn, arg, msg):
|
||||
# compares ppf for multiple argsets.
|
||||
num_repeats = 5
|
||||
args = [] * num_repeats
|
||||
if arg:
|
||||
args = [np.array([_] * num_repeats) for _ in arg]
|
||||
|
||||
median = distfn.ppf(0.5, *arg)
|
||||
medians = distfn.ppf(0.5, *args)
|
||||
msg += " - ppf multiple"
|
||||
npt.assert_almost_equal(medians, [median] * num_repeats, decimal=7, err_msg=msg)
|
||||
|
||||
|
||||
def check_distribution_rvs(dist, args, alpha, rvs):
|
||||
# dist is either a cdf function or name of a distribution in scipy.stats.
|
||||
# args are the args for scipy.stats.dist(*args)
|
||||
# alpha is a significance level, ~0.01
|
||||
# rvs is array_like of random variables
|
||||
# test from scipy.stats.tests
|
||||
# this version reuses existing random variables
|
||||
D, pval = stats.kstest(rvs, dist, args=args, N=1000)
|
||||
if (pval < alpha):
|
||||
# The rvs passed in failed the K-S test, which _could_ happen
|
||||
# but is unlikely if alpha is small enough.
|
||||
# Repeat the the test with a new sample of rvs.
|
||||
# Generate 1000 rvs, perform a K-S test that the new sample of rvs
|
||||
# are distributed according to the distribution.
|
||||
D, pval = stats.kstest(dist, dist, args=args, N=1000)
|
||||
npt.assert_(pval > alpha, "D = " + str(D) + "; pval = " + str(pval) +
|
||||
"; alpha = " + str(alpha) + "\nargs = " + str(args))
|
||||
|
||||
|
||||
def check_vecentropy(distfn, args):
|
||||
npt.assert_equal(distfn.vecentropy(*args), distfn._entropy(*args))
|
||||
|
||||
|
||||
def check_loc_scale(distfn, arg, m, v, msg):
|
||||
# Make `loc` and `scale` arrays to catch bugs like gh-13580 where
|
||||
# `loc` and `scale` arrays improperly broadcast with shapes.
|
||||
loc, scale = np.array([10.0, 20.0]), np.array([10.0, 20.0])
|
||||
mt, vt = distfn.stats(loc=loc, scale=scale, *arg)
|
||||
npt.assert_allclose(m*scale + loc, mt)
|
||||
npt.assert_allclose(v*scale*scale, vt)
|
||||
|
||||
|
||||
def check_ppf_private(distfn, arg, msg):
|
||||
# fails by design for truncnorm self.nb not defined
|
||||
ppfs = distfn._ppf(np.array([0.1, 0.5, 0.9]), *arg)
|
||||
npt.assert_(not np.any(np.isnan(ppfs)), msg + 'ppf private is nan')
|
||||
|
||||
|
||||
def check_retrieving_support(distfn, args):
|
||||
loc, scale = 1, 2
|
||||
supp = distfn.support(*args)
|
||||
supp_loc_scale = distfn.support(*args, loc=loc, scale=scale)
|
||||
npt.assert_almost_equal(np.array(supp)*scale + loc,
|
||||
np.array(supp_loc_scale))
|
||||
|
||||
|
||||
def check_fit_args(distfn, arg, rvs, method):
|
||||
with np.errstate(all='ignore'), npt.suppress_warnings() as sup:
|
||||
sup.filter(category=RuntimeWarning,
|
||||
message="The shape parameter of the erlang")
|
||||
sup.filter(category=RuntimeWarning,
|
||||
message="floating point number truncated")
|
||||
vals = distfn.fit(rvs, method=method)
|
||||
vals2 = distfn.fit(rvs, optimizer='powell', method=method)
|
||||
# Only check the length of the return; accuracy tested in test_fit.py
|
||||
npt.assert_(len(vals) == 2+len(arg))
|
||||
npt.assert_(len(vals2) == 2+len(arg))
|
||||
|
||||
|
||||
def check_fit_args_fix(distfn, arg, rvs, method):
|
||||
with np.errstate(all='ignore'), npt.suppress_warnings() as sup:
|
||||
sup.filter(category=RuntimeWarning,
|
||||
message="The shape parameter of the erlang")
|
||||
|
||||
vals = distfn.fit(rvs, floc=0, method=method)
|
||||
vals2 = distfn.fit(rvs, fscale=1, method=method)
|
||||
npt.assert_(len(vals) == 2+len(arg))
|
||||
npt.assert_(vals[-2] == 0)
|
||||
npt.assert_(vals2[-1] == 1)
|
||||
npt.assert_(len(vals2) == 2+len(arg))
|
||||
if len(arg) > 0:
|
||||
vals3 = distfn.fit(rvs, f0=arg[0], method=method)
|
||||
npt.assert_(len(vals3) == 2+len(arg))
|
||||
npt.assert_(vals3[0] == arg[0])
|
||||
if len(arg) > 1:
|
||||
vals4 = distfn.fit(rvs, f1=arg[1], method=method)
|
||||
npt.assert_(len(vals4) == 2+len(arg))
|
||||
npt.assert_(vals4[1] == arg[1])
|
||||
if len(arg) > 2:
|
||||
vals5 = distfn.fit(rvs, f2=arg[2], method=method)
|
||||
npt.assert_(len(vals5) == 2+len(arg))
|
||||
npt.assert_(vals5[2] == arg[2])
|
||||
|
||||
|
||||
@pytest.mark.parametrize('method', ['pdf', 'logpdf', 'cdf', 'logcdf',
|
||||
'sf', 'logsf', 'ppf', 'isf'])
|
||||
@pytest.mark.parametrize('distname, args', distcont)
|
||||
def test_methods_with_lists(method, distname, args):
|
||||
# Test that the continuous distributions can accept Python lists
|
||||
# as arguments.
|
||||
dist = getattr(stats, distname)
|
||||
f = getattr(dist, method)
|
||||
if distname == 'invweibull' and method.startswith('log'):
|
||||
x = [1.5, 2]
|
||||
else:
|
||||
x = [0.1, 0.2]
|
||||
|
||||
shape2 = [[a]*2 for a in args]
|
||||
loc = [0, 0.1]
|
||||
scale = [1, 1.01]
|
||||
result = f(x, *shape2, loc=loc, scale=scale)
|
||||
npt.assert_allclose(result,
|
||||
[f(*v) for v in zip(x, *shape2, loc, scale)],
|
||||
rtol=1e-14, atol=5e-14)
|
||||
|
||||
|
||||
def test_burr_fisk_moment_gh13234_regression():
|
||||
vals0 = stats.burr.moment(1, 5, 4)
|
||||
assert isinstance(vals0, float)
|
||||
|
||||
vals1 = stats.fisk.moment(1, 8)
|
||||
assert isinstance(vals1, float)
|
||||
|
||||
|
||||
def test_moments_with_array_gh12192_regression():
|
||||
# array loc and scalar scale
|
||||
vals0 = stats.norm.moment(n=1, loc=np.array([1, 2, 3]), scale=1)
|
||||
expected0 = np.array([1., 2., 3.])
|
||||
npt.assert_equal(vals0, expected0)
|
||||
|
||||
# array loc and invalid scalar scale
|
||||
vals1 = stats.norm.moment(n=1, loc=np.array([1, 2, 3]), scale=-1)
|
||||
expected1 = np.array([np.nan, np.nan, np.nan])
|
||||
npt.assert_equal(vals1, expected1)
|
||||
|
||||
# array loc and array scale with invalid entries
|
||||
vals2 = stats.norm.moment(n=1, loc=np.array([1, 2, 3]), scale=[-3, 1, 0])
|
||||
expected2 = np.array([np.nan, 2., np.nan])
|
||||
npt.assert_equal(vals2, expected2)
|
||||
|
||||
# (loc == 0) & (scale < 0)
|
||||
vals3 = stats.norm.moment(n=2, loc=0, scale=-4)
|
||||
expected3 = np.nan
|
||||
npt.assert_equal(vals3, expected3)
|
||||
assert isinstance(vals3, expected3.__class__)
|
||||
|
||||
# array loc with 0 entries and scale with invalid entries
|
||||
vals4 = stats.norm.moment(n=2, loc=[1, 0, 2], scale=[3, -4, -5])
|
||||
expected4 = np.array([10., np.nan, np.nan])
|
||||
npt.assert_equal(vals4, expected4)
|
||||
|
||||
# all(loc == 0) & (array scale with invalid entries)
|
||||
vals5 = stats.norm.moment(n=2, loc=[0, 0, 0], scale=[5., -2, 100.])
|
||||
expected5 = np.array([25., np.nan, 10000.])
|
||||
npt.assert_equal(vals5, expected5)
|
||||
|
||||
# all( (loc == 0) & (scale < 0) )
|
||||
vals6 = stats.norm.moment(n=2, loc=[0, 0, 0], scale=[-5., -2, -100.])
|
||||
expected6 = np.array([np.nan, np.nan, np.nan])
|
||||
npt.assert_equal(vals6, expected6)
|
||||
|
||||
# scalar args, loc, and scale
|
||||
vals7 = stats.chi.moment(n=2, df=1, loc=0, scale=0)
|
||||
expected7 = np.nan
|
||||
npt.assert_equal(vals7, expected7)
|
||||
assert isinstance(vals7, expected7.__class__)
|
||||
|
||||
# array args, scalar loc, and scalar scale
|
||||
vals8 = stats.chi.moment(n=2, df=[1, 2, 3], loc=0, scale=0)
|
||||
expected8 = np.array([np.nan, np.nan, np.nan])
|
||||
npt.assert_equal(vals8, expected8)
|
||||
|
||||
# array args, array loc, and array scale
|
||||
vals9 = stats.chi.moment(n=2, df=[1, 2, 3], loc=[1., 0., 2.],
|
||||
scale=[1., -3., 0.])
|
||||
expected9 = np.array([3.59576912, np.nan, np.nan])
|
||||
npt.assert_allclose(vals9, expected9, rtol=1e-8)
|
||||
|
||||
# (n > 4), all(loc != 0), and all(scale != 0)
|
||||
vals10 = stats.norm.moment(5, [1., 2.], [1., 2.])
|
||||
expected10 = np.array([26., 832.])
|
||||
npt.assert_allclose(vals10, expected10, rtol=1e-13)
|
||||
|
||||
# test broadcasting and more
|
||||
a = [-1.1, 0, 1, 2.2, np.pi]
|
||||
b = [-1.1, 0, 1, 2.2, np.pi]
|
||||
loc = [-1.1, 0, np.sqrt(2)]
|
||||
scale = [-2.1, 0, 1, 2.2, np.pi]
|
||||
|
||||
a = np.array(a).reshape((-1, 1, 1, 1))
|
||||
b = np.array(b).reshape((-1, 1, 1))
|
||||
loc = np.array(loc).reshape((-1, 1))
|
||||
scale = np.array(scale)
|
||||
|
||||
vals11 = stats.beta.moment(n=2, a=a, b=b, loc=loc, scale=scale)
|
||||
|
||||
a, b, loc, scale = np.broadcast_arrays(a, b, loc, scale)
|
||||
|
||||
for i in np.ndenumerate(a):
|
||||
with np.errstate(invalid='ignore', divide='ignore'):
|
||||
i = i[0] # just get the index
|
||||
# check against same function with scalar input
|
||||
expected = stats.beta.moment(n=2, a=a[i], b=b[i],
|
||||
loc=loc[i], scale=scale[i])
|
||||
np.testing.assert_equal(vals11[i], expected)
|
||||
|
||||
|
||||
def test_broadcasting_in_moments_gh12192_regression():
|
||||
vals0 = stats.norm.moment(n=1, loc=np.array([1, 2, 3]), scale=[[1]])
|
||||
expected0 = np.array([[1., 2., 3.]])
|
||||
npt.assert_equal(vals0, expected0)
|
||||
assert vals0.shape == expected0.shape
|
||||
|
||||
vals1 = stats.norm.moment(n=1, loc=np.array([[1], [2], [3]]),
|
||||
scale=[1, 2, 3])
|
||||
expected1 = np.array([[1., 1., 1.], [2., 2., 2.], [3., 3., 3.]])
|
||||
npt.assert_equal(vals1, expected1)
|
||||
assert vals1.shape == expected1.shape
|
||||
|
||||
vals2 = stats.chi.moment(n=1, df=[1., 2., 3.], loc=0., scale=1.)
|
||||
expected2 = np.array([0.79788456, 1.25331414, 1.59576912])
|
||||
npt.assert_allclose(vals2, expected2, rtol=1e-8)
|
||||
assert vals2.shape == expected2.shape
|
||||
|
||||
vals3 = stats.chi.moment(n=1, df=[[1.], [2.], [3.]], loc=[0., 1., 2.],
|
||||
scale=[-1., 0., 3.])
|
||||
expected3 = np.array([[np.nan, np.nan, 4.39365368],
|
||||
[np.nan, np.nan, 5.75994241],
|
||||
[np.nan, np.nan, 6.78730736]])
|
||||
npt.assert_allclose(vals3, expected3, rtol=1e-8)
|
||||
assert vals3.shape == expected3.shape
|
||||
@@ -0,0 +1,110 @@
|
||||
import pytest
|
||||
import numpy as np
|
||||
from numpy.testing import assert_array_equal
|
||||
from scipy.stats.contingency import crosstab
|
||||
|
||||
|
||||
@pytest.mark.parametrize('sparse', [False, True])
|
||||
def test_crosstab_basic(sparse):
|
||||
a = [0, 0, 9, 9, 0, 0, 9]
|
||||
b = [2, 1, 3, 1, 2, 3, 3]
|
||||
expected_avals = [0, 9]
|
||||
expected_bvals = [1, 2, 3]
|
||||
expected_count = np.array([[1, 2, 1],
|
||||
[1, 0, 2]])
|
||||
(avals, bvals), count = crosstab(a, b, sparse=sparse)
|
||||
assert_array_equal(avals, expected_avals)
|
||||
assert_array_equal(bvals, expected_bvals)
|
||||
if sparse:
|
||||
assert_array_equal(count.A, expected_count)
|
||||
else:
|
||||
assert_array_equal(count, expected_count)
|
||||
|
||||
|
||||
def test_crosstab_basic_1d():
|
||||
# Verify that a single input sequence works as expected.
|
||||
x = [1, 2, 3, 1, 2, 3, 3]
|
||||
expected_xvals = [1, 2, 3]
|
||||
expected_count = np.array([2, 2, 3])
|
||||
(xvals,), count = crosstab(x)
|
||||
assert_array_equal(xvals, expected_xvals)
|
||||
assert_array_equal(count, expected_count)
|
||||
|
||||
|
||||
def test_crosstab_basic_3d():
|
||||
# Verify the function for three input sequences.
|
||||
a = 'a'
|
||||
b = 'b'
|
||||
x = [0, 0, 9, 9, 0, 0, 9, 9]
|
||||
y = [a, a, a, a, b, b, b, a]
|
||||
z = [1, 2, 3, 1, 2, 3, 3, 1]
|
||||
expected_xvals = [0, 9]
|
||||
expected_yvals = [a, b]
|
||||
expected_zvals = [1, 2, 3]
|
||||
expected_count = np.array([[[1, 1, 0],
|
||||
[0, 1, 1]],
|
||||
[[2, 0, 1],
|
||||
[0, 0, 1]]])
|
||||
(xvals, yvals, zvals), count = crosstab(x, y, z)
|
||||
assert_array_equal(xvals, expected_xvals)
|
||||
assert_array_equal(yvals, expected_yvals)
|
||||
assert_array_equal(zvals, expected_zvals)
|
||||
assert_array_equal(count, expected_count)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('sparse', [False, True])
|
||||
def test_crosstab_levels(sparse):
|
||||
a = [0, 0, 9, 9, 0, 0, 9]
|
||||
b = [1, 2, 3, 1, 2, 3, 3]
|
||||
expected_avals = [0, 9]
|
||||
expected_bvals = [0, 1, 2, 3]
|
||||
expected_count = np.array([[0, 1, 2, 1],
|
||||
[0, 1, 0, 2]])
|
||||
(avals, bvals), count = crosstab(a, b, levels=[None, [0, 1, 2, 3]],
|
||||
sparse=sparse)
|
||||
assert_array_equal(avals, expected_avals)
|
||||
assert_array_equal(bvals, expected_bvals)
|
||||
if sparse:
|
||||
assert_array_equal(count.A, expected_count)
|
||||
else:
|
||||
assert_array_equal(count, expected_count)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('sparse', [False, True])
|
||||
def test_crosstab_extra_levels(sparse):
|
||||
# The pair of values (-1, 3) will be ignored, because we explicitly
|
||||
# request the counted `a` values to be [0, 9].
|
||||
a = [0, 0, 9, 9, 0, 0, 9, -1]
|
||||
b = [1, 2, 3, 1, 2, 3, 3, 3]
|
||||
expected_avals = [0, 9]
|
||||
expected_bvals = [0, 1, 2, 3]
|
||||
expected_count = np.array([[0, 1, 2, 1],
|
||||
[0, 1, 0, 2]])
|
||||
(avals, bvals), count = crosstab(a, b, levels=[[0, 9], [0, 1, 2, 3]],
|
||||
sparse=sparse)
|
||||
assert_array_equal(avals, expected_avals)
|
||||
assert_array_equal(bvals, expected_bvals)
|
||||
if sparse:
|
||||
assert_array_equal(count.A, expected_count)
|
||||
else:
|
||||
assert_array_equal(count, expected_count)
|
||||
|
||||
|
||||
def test_validation_at_least_one():
|
||||
with pytest.raises(TypeError, match='At least one'):
|
||||
crosstab()
|
||||
|
||||
|
||||
def test_validation_same_lengths():
|
||||
with pytest.raises(ValueError, match='must have the same length'):
|
||||
crosstab([1, 2], [1, 2, 3, 4])
|
||||
|
||||
|
||||
def test_validation_sparse_only_two_args():
|
||||
with pytest.raises(ValueError, match='only two input sequences'):
|
||||
crosstab([0, 1, 1], [8, 8, 9], [1, 3, 3], sparse=True)
|
||||
|
||||
|
||||
def test_validation_len_levels_matches_args():
|
||||
with pytest.raises(ValueError, match='number of input sequences'):
|
||||
crosstab([0, 1, 1], [8, 8, 9], levels=([0, 1, 2, 3],))
|
||||
@@ -0,0 +1,317 @@
|
||||
import numpy.testing as npt
|
||||
import numpy as np
|
||||
import pytest
|
||||
|
||||
from scipy import stats
|
||||
from .common_tests import (check_normalization, check_moment, check_mean_expect,
|
||||
check_var_expect, check_skew_expect,
|
||||
check_kurt_expect, check_entropy,
|
||||
check_private_entropy, check_edge_support,
|
||||
check_named_args, check_random_state_property,
|
||||
check_pickling, check_rvs_broadcast, check_freezing)
|
||||
from scipy.stats._distr_params import distdiscrete, invdistdiscrete
|
||||
|
||||
vals = ([1, 2, 3, 4], [0.1, 0.2, 0.3, 0.4])
|
||||
distdiscrete += [[stats.rv_discrete(values=vals), ()]]
|
||||
|
||||
# For these distributions, test_discrete_basic only runs with test mode full
|
||||
distslow = {'zipfian', 'nhypergeom'}
|
||||
|
||||
|
||||
def cases_test_discrete_basic():
|
||||
seen = set()
|
||||
for distname, arg in distdiscrete:
|
||||
if distname in distslow:
|
||||
yield pytest.param(distname, arg, distname, marks=pytest.mark.slow)
|
||||
else:
|
||||
yield distname, arg, distname not in seen
|
||||
seen.add(distname)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('distname,arg,first_case', cases_test_discrete_basic())
|
||||
def test_discrete_basic(distname, arg, first_case):
|
||||
try:
|
||||
distfn = getattr(stats, distname)
|
||||
except TypeError:
|
||||
distfn = distname
|
||||
distname = 'sample distribution'
|
||||
np.random.seed(9765456)
|
||||
rvs = distfn.rvs(size=2000, *arg)
|
||||
supp = np.unique(rvs)
|
||||
m, v = distfn.stats(*arg)
|
||||
check_cdf_ppf(distfn, arg, supp, distname + ' cdf_ppf')
|
||||
|
||||
check_pmf_cdf(distfn, arg, distname)
|
||||
check_oth(distfn, arg, supp, distname + ' oth')
|
||||
check_edge_support(distfn, arg)
|
||||
|
||||
alpha = 0.01
|
||||
check_discrete_chisquare(distfn, arg, rvs, alpha,
|
||||
distname + ' chisquare')
|
||||
|
||||
if first_case:
|
||||
locscale_defaults = (0,)
|
||||
meths = [distfn.pmf, distfn.logpmf, distfn.cdf, distfn.logcdf,
|
||||
distfn.logsf]
|
||||
# make sure arguments are within support
|
||||
# for some distributions, this needs to be overridden
|
||||
spec_k = {'randint': 11, 'hypergeom': 4, 'bernoulli': 0,
|
||||
'nchypergeom_wallenius': 6}
|
||||
k = spec_k.get(distname, 1)
|
||||
check_named_args(distfn, k, arg, locscale_defaults, meths)
|
||||
if distname != 'sample distribution':
|
||||
check_scale_docstring(distfn)
|
||||
check_random_state_property(distfn, arg)
|
||||
check_pickling(distfn, arg)
|
||||
check_freezing(distfn, arg)
|
||||
|
||||
# Entropy
|
||||
check_entropy(distfn, arg, distname)
|
||||
if distfn.__class__._entropy != stats.rv_discrete._entropy:
|
||||
check_private_entropy(distfn, arg, stats.rv_discrete)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('distname,arg', distdiscrete)
|
||||
def test_moments(distname, arg):
|
||||
try:
|
||||
distfn = getattr(stats, distname)
|
||||
except TypeError:
|
||||
distfn = distname
|
||||
distname = 'sample distribution'
|
||||
m, v, s, k = distfn.stats(*arg, moments='mvsk')
|
||||
check_normalization(distfn, arg, distname)
|
||||
|
||||
# compare `stats` and `moment` methods
|
||||
check_moment(distfn, arg, m, v, distname)
|
||||
check_mean_expect(distfn, arg, m, distname)
|
||||
check_var_expect(distfn, arg, m, v, distname)
|
||||
check_skew_expect(distfn, arg, m, v, s, distname)
|
||||
if distname not in ['zipf', 'yulesimon']:
|
||||
check_kurt_expect(distfn, arg, m, v, k, distname)
|
||||
|
||||
# frozen distr moments
|
||||
check_moment_frozen(distfn, arg, m, 1)
|
||||
check_moment_frozen(distfn, arg, v+m*m, 2)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('dist,shape_args', distdiscrete)
|
||||
def test_rvs_broadcast(dist, shape_args):
|
||||
# If shape_only is True, it means the _rvs method of the
|
||||
# distribution uses more than one random number to generate a random
|
||||
# variate. That means the result of using rvs with broadcasting or
|
||||
# with a nontrivial size will not necessarily be the same as using the
|
||||
# numpy.vectorize'd version of rvs(), so we can only compare the shapes
|
||||
# of the results, not the values.
|
||||
# Whether or not a distribution is in the following list is an
|
||||
# implementation detail of the distribution, not a requirement. If
|
||||
# the implementation the rvs() method of a distribution changes, this
|
||||
# test might also have to be changed.
|
||||
shape_only = dist in ['betabinom', 'skellam', 'yulesimon', 'dlaplace',
|
||||
'nchypergeom_fisher', 'nchypergeom_wallenius']
|
||||
|
||||
try:
|
||||
distfunc = getattr(stats, dist)
|
||||
except TypeError:
|
||||
distfunc = dist
|
||||
dist = 'rv_discrete(values=(%r, %r))' % (dist.xk, dist.pk)
|
||||
loc = np.zeros(2)
|
||||
nargs = distfunc.numargs
|
||||
allargs = []
|
||||
bshape = []
|
||||
# Generate shape parameter arguments...
|
||||
for k in range(nargs):
|
||||
shp = (k + 3,) + (1,)*(k + 1)
|
||||
param_val = shape_args[k]
|
||||
allargs.append(np.full(shp, param_val))
|
||||
bshape.insert(0, shp[0])
|
||||
allargs.append(loc)
|
||||
bshape.append(loc.size)
|
||||
# bshape holds the expected shape when loc, scale, and the shape
|
||||
# parameters are all broadcast together.
|
||||
check_rvs_broadcast(distfunc, dist, allargs, bshape, shape_only, [np.int_])
|
||||
|
||||
|
||||
@pytest.mark.parametrize('dist,args', distdiscrete)
|
||||
def test_ppf_with_loc(dist, args):
|
||||
try:
|
||||
distfn = getattr(stats, dist)
|
||||
except TypeError:
|
||||
distfn = dist
|
||||
#check with a negative, no and positive relocation.
|
||||
np.random.seed(1942349)
|
||||
re_locs = [np.random.randint(-10, -1), 0, np.random.randint(1, 10)]
|
||||
_a, _b = distfn.support(*args)
|
||||
for loc in re_locs:
|
||||
npt.assert_array_equal(
|
||||
[_a-1+loc, _b+loc],
|
||||
[distfn.ppf(0.0, *args, loc=loc), distfn.ppf(1.0, *args, loc=loc)]
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('dist, args', distdiscrete)
|
||||
def test_isf_with_loc(dist, args):
|
||||
try:
|
||||
distfn = getattr(stats, dist)
|
||||
except TypeError:
|
||||
distfn = dist
|
||||
# check with a negative, no and positive relocation.
|
||||
np.random.seed(1942349)
|
||||
re_locs = [np.random.randint(-10, -1), 0, np.random.randint(1, 10)]
|
||||
_a, _b = distfn.support(*args)
|
||||
for loc in re_locs:
|
||||
expected = _b + loc, _a - 1 + loc
|
||||
res = distfn.isf(0., *args, loc=loc), distfn.isf(1., *args, loc=loc)
|
||||
npt.assert_array_equal(expected, res)
|
||||
# test broadcasting behaviour
|
||||
re_locs = [np.random.randint(-10, -1, size=(5, 3)),
|
||||
np.zeros((5, 3)),
|
||||
np.random.randint(1, 10, size=(5, 3))]
|
||||
_a, _b = distfn.support(*args)
|
||||
for loc in re_locs:
|
||||
expected = _b + loc, _a - 1 + loc
|
||||
res = distfn.isf(0., *args, loc=loc), distfn.isf(1., *args, loc=loc)
|
||||
npt.assert_array_equal(expected, res)
|
||||
|
||||
|
||||
def check_cdf_ppf(distfn, arg, supp, msg):
|
||||
# cdf is a step function, and ppf(q) = min{k : cdf(k) >= q, k integer}
|
||||
npt.assert_array_equal(distfn.ppf(distfn.cdf(supp, *arg), *arg),
|
||||
supp, msg + '-roundtrip')
|
||||
npt.assert_array_equal(distfn.ppf(distfn.cdf(supp, *arg) - 1e-8, *arg),
|
||||
supp, msg + '-roundtrip')
|
||||
|
||||
if not hasattr(distfn, 'xk'):
|
||||
_a, _b = distfn.support(*arg)
|
||||
supp1 = supp[supp < _b]
|
||||
npt.assert_array_equal(distfn.ppf(distfn.cdf(supp1, *arg) + 1e-8, *arg),
|
||||
supp1 + distfn.inc, msg + ' ppf-cdf-next')
|
||||
# -1e-8 could cause an error if pmf < 1e-8
|
||||
|
||||
|
||||
def check_pmf_cdf(distfn, arg, distname):
|
||||
if hasattr(distfn, 'xk'):
|
||||
index = distfn.xk
|
||||
else:
|
||||
startind = int(distfn.ppf(0.01, *arg) - 1)
|
||||
index = list(range(startind, startind + 10))
|
||||
cdfs = distfn.cdf(index, *arg)
|
||||
pmfs_cum = distfn.pmf(index, *arg).cumsum()
|
||||
|
||||
atol, rtol = 1e-10, 1e-10
|
||||
if distname == 'skellam': # ncx2 accuracy
|
||||
atol, rtol = 1e-5, 1e-5
|
||||
npt.assert_allclose(cdfs - cdfs[0], pmfs_cum - pmfs_cum[0],
|
||||
atol=atol, rtol=rtol)
|
||||
|
||||
|
||||
def check_moment_frozen(distfn, arg, m, k):
|
||||
npt.assert_allclose(distfn(*arg).moment(k), m,
|
||||
atol=1e-10, rtol=1e-10)
|
||||
|
||||
|
||||
def check_oth(distfn, arg, supp, msg):
|
||||
# checking other methods of distfn
|
||||
npt.assert_allclose(distfn.sf(supp, *arg), 1. - distfn.cdf(supp, *arg),
|
||||
atol=1e-10, rtol=1e-10)
|
||||
|
||||
q = np.linspace(0.01, 0.99, 20)
|
||||
npt.assert_allclose(distfn.isf(q, *arg), distfn.ppf(1. - q, *arg),
|
||||
atol=1e-10, rtol=1e-10)
|
||||
|
||||
median_sf = distfn.isf(0.5, *arg)
|
||||
npt.assert_(distfn.sf(median_sf - 1, *arg) > 0.5)
|
||||
npt.assert_(distfn.cdf(median_sf + 1, *arg) > 0.5)
|
||||
|
||||
|
||||
def check_discrete_chisquare(distfn, arg, rvs, alpha, msg):
|
||||
"""Perform chisquare test for random sample of a discrete distribution
|
||||
|
||||
Parameters
|
||||
----------
|
||||
distname : string
|
||||
name of distribution function
|
||||
arg : sequence
|
||||
parameters of distribution
|
||||
alpha : float
|
||||
significance level, threshold for p-value
|
||||
|
||||
Returns
|
||||
-------
|
||||
result : bool
|
||||
0 if test passes, 1 if test fails
|
||||
|
||||
"""
|
||||
wsupp = 0.05
|
||||
|
||||
# construct intervals with minimum mass `wsupp`.
|
||||
# intervals are left-half-open as in a cdf difference
|
||||
_a, _b = distfn.support(*arg)
|
||||
lo = int(max(_a, -1000))
|
||||
high = int(min(_b, 1000)) + 1
|
||||
distsupport = range(lo, high)
|
||||
last = 0
|
||||
distsupp = [lo]
|
||||
distmass = []
|
||||
for ii in distsupport:
|
||||
current = distfn.cdf(ii, *arg)
|
||||
if current - last >= wsupp - 1e-14:
|
||||
distsupp.append(ii)
|
||||
distmass.append(current - last)
|
||||
last = current
|
||||
if current > (1 - wsupp):
|
||||
break
|
||||
if distsupp[-1] < _b:
|
||||
distsupp.append(_b)
|
||||
distmass.append(1 - last)
|
||||
distsupp = np.array(distsupp)
|
||||
distmass = np.array(distmass)
|
||||
|
||||
# convert intervals to right-half-open as required by histogram
|
||||
histsupp = distsupp + 1e-8
|
||||
histsupp[0] = _a
|
||||
|
||||
# find sample frequencies and perform chisquare test
|
||||
freq, hsupp = np.histogram(rvs, histsupp)
|
||||
chis, pval = stats.chisquare(np.array(freq), len(rvs)*distmass)
|
||||
|
||||
npt.assert_(pval > alpha,
|
||||
'chisquare - test for %s at arg = %s with pval = %s' %
|
||||
(msg, str(arg), str(pval)))
|
||||
|
||||
|
||||
def check_scale_docstring(distfn):
|
||||
if distfn.__doc__ is not None:
|
||||
# Docstrings can be stripped if interpreter is run with -OO
|
||||
npt.assert_('scale' not in distfn.__doc__)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('method', ['pmf', 'logpmf', 'cdf', 'logcdf',
|
||||
'sf', 'logsf', 'ppf', 'isf'])
|
||||
@pytest.mark.parametrize('distname, args', distdiscrete)
|
||||
def test_methods_with_lists(method, distname, args):
|
||||
# Test that the discrete distributions can accept Python lists
|
||||
# as arguments.
|
||||
try:
|
||||
dist = getattr(stats, distname)
|
||||
except TypeError:
|
||||
return
|
||||
if method in ['ppf', 'isf']:
|
||||
z = [0.1, 0.2]
|
||||
else:
|
||||
z = [0, 1]
|
||||
p2 = [[p]*2 for p in args]
|
||||
loc = [0, 1]
|
||||
result = dist.pmf(z, *p2, loc=loc)
|
||||
npt.assert_allclose(result,
|
||||
[dist.pmf(*v) for v in zip(z, *p2, loc)],
|
||||
rtol=1e-15, atol=1e-15)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('distname, args', invdistdiscrete)
|
||||
def test_cdf_gh13280_regression(distname, args):
|
||||
# Test for nan output when shape parameters are invalid
|
||||
dist = getattr(stats, distname)
|
||||
x = np.arange(-2, 15)
|
||||
vals = dist.cdf(x, *args)
|
||||
expected = np.nan
|
||||
npt.assert_equal(vals, expected)
|
||||
@@ -0,0 +1,546 @@
|
||||
import pytest
|
||||
from scipy.stats import (betabinom, hypergeom, nhypergeom, bernoulli,
|
||||
boltzmann, skellam, zipf, zipfian, binom, nbinom,
|
||||
nchypergeom_fisher, nchypergeom_wallenius, randint)
|
||||
|
||||
import numpy as np
|
||||
from numpy.testing import (
|
||||
assert_almost_equal, assert_equal, assert_allclose, suppress_warnings
|
||||
)
|
||||
from scipy.special import binom as special_binom
|
||||
from scipy.optimize import root_scalar
|
||||
from scipy.integrate import quad
|
||||
|
||||
|
||||
# The expected values were computed with Wolfram Alpha, using
|
||||
# the expression CDF[HypergeometricDistribution[N, n, M], k].
|
||||
@pytest.mark.parametrize('k, M, n, N, expected, rtol',
|
||||
[(3, 10, 4, 5,
|
||||
0.9761904761904762, 1e-15),
|
||||
(107, 10000, 3000, 215,
|
||||
0.9999999997226765, 1e-15),
|
||||
(10, 10000, 3000, 215,
|
||||
2.681682217692179e-21, 5e-11)])
|
||||
def test_hypergeom_cdf(k, M, n, N, expected, rtol):
|
||||
p = hypergeom.cdf(k, M, n, N)
|
||||
assert_allclose(p, expected, rtol=rtol)
|
||||
|
||||
|
||||
# The expected values were computed with Wolfram Alpha, using
|
||||
# the expression SurvivalFunction[HypergeometricDistribution[N, n, M], k].
|
||||
@pytest.mark.parametrize('k, M, n, N, expected, rtol',
|
||||
[(25, 10000, 3000, 215,
|
||||
0.9999999999052958, 1e-15),
|
||||
(125, 10000, 3000, 215,
|
||||
1.4416781705752128e-18, 5e-11)])
|
||||
def test_hypergeom_sf(k, M, n, N, expected, rtol):
|
||||
p = hypergeom.sf(k, M, n, N)
|
||||
assert_allclose(p, expected, rtol=rtol)
|
||||
|
||||
|
||||
def test_hypergeom_logpmf():
|
||||
# symmetries test
|
||||
# f(k,N,K,n) = f(n-k,N,N-K,n) = f(K-k,N,K,N-n) = f(k,N,n,K)
|
||||
k = 5
|
||||
N = 50
|
||||
K = 10
|
||||
n = 5
|
||||
logpmf1 = hypergeom.logpmf(k, N, K, n)
|
||||
logpmf2 = hypergeom.logpmf(n - k, N, N - K, n)
|
||||
logpmf3 = hypergeom.logpmf(K - k, N, K, N - n)
|
||||
logpmf4 = hypergeom.logpmf(k, N, n, K)
|
||||
assert_almost_equal(logpmf1, logpmf2, decimal=12)
|
||||
assert_almost_equal(logpmf1, logpmf3, decimal=12)
|
||||
assert_almost_equal(logpmf1, logpmf4, decimal=12)
|
||||
|
||||
# test related distribution
|
||||
# Bernoulli distribution if n = 1
|
||||
k = 1
|
||||
N = 10
|
||||
K = 7
|
||||
n = 1
|
||||
hypergeom_logpmf = hypergeom.logpmf(k, N, K, n)
|
||||
bernoulli_logpmf = bernoulli.logpmf(k, K/N)
|
||||
assert_almost_equal(hypergeom_logpmf, bernoulli_logpmf, decimal=12)
|
||||
|
||||
|
||||
def test_nhypergeom_pmf():
|
||||
# test with hypergeom
|
||||
M, n, r = 45, 13, 8
|
||||
k = 6
|
||||
NHG = nhypergeom.pmf(k, M, n, r)
|
||||
HG = hypergeom.pmf(k, M, n, k+r-1) * (M - n - (r-1)) / (M - (k+r-1))
|
||||
assert_allclose(HG, NHG, rtol=1e-10)
|
||||
|
||||
|
||||
def test_nhypergeom_pmfcdf():
|
||||
# test pmf and cdf with arbitrary values.
|
||||
M = 8
|
||||
n = 3
|
||||
r = 4
|
||||
support = np.arange(n+1)
|
||||
pmf = nhypergeom.pmf(support, M, n, r)
|
||||
cdf = nhypergeom.cdf(support, M, n, r)
|
||||
assert_allclose(pmf, [1/14, 3/14, 5/14, 5/14], rtol=1e-13)
|
||||
assert_allclose(cdf, [1/14, 4/14, 9/14, 1.0], rtol=1e-13)
|
||||
|
||||
|
||||
def test_nhypergeom_r0():
|
||||
# test with `r = 0`.
|
||||
M = 10
|
||||
n = 3
|
||||
r = 0
|
||||
pmf = nhypergeom.pmf([[0, 1, 2, 0], [1, 2, 0, 3]], M, n, r)
|
||||
assert_allclose(pmf, [[1, 0, 0, 1], [0, 0, 1, 0]], rtol=1e-13)
|
||||
|
||||
|
||||
def test_nhypergeom_rvs_shape():
|
||||
# Check that when given a size with more dimensions than the
|
||||
# dimensions of the broadcast parameters, rvs returns an array
|
||||
# with the correct shape.
|
||||
x = nhypergeom.rvs(22, [7, 8, 9], [[12], [13]], size=(5, 1, 2, 3))
|
||||
assert x.shape == (5, 1, 2, 3)
|
||||
|
||||
|
||||
def test_nhypergeom_accuracy():
|
||||
# Check that nhypergeom.rvs post-gh-13431 gives the same values as
|
||||
# inverse transform sampling
|
||||
np.random.seed(0)
|
||||
x = nhypergeom.rvs(22, 7, 11, size=100)
|
||||
np.random.seed(0)
|
||||
p = np.random.uniform(size=100)
|
||||
y = nhypergeom.ppf(p, 22, 7, 11)
|
||||
assert_equal(x, y)
|
||||
|
||||
|
||||
def test_boltzmann_upper_bound():
|
||||
k = np.arange(-3, 5)
|
||||
|
||||
N = 1
|
||||
p = boltzmann.pmf(k, 0.123, N)
|
||||
expected = k == 0
|
||||
assert_equal(p, expected)
|
||||
|
||||
lam = np.log(2)
|
||||
N = 3
|
||||
p = boltzmann.pmf(k, lam, N)
|
||||
expected = [0, 0, 0, 4/7, 2/7, 1/7, 0, 0]
|
||||
assert_allclose(p, expected, rtol=1e-13)
|
||||
|
||||
c = boltzmann.cdf(k, lam, N)
|
||||
expected = [0, 0, 0, 4/7, 6/7, 1, 1, 1]
|
||||
assert_allclose(c, expected, rtol=1e-13)
|
||||
|
||||
|
||||
def test_betabinom_a_and_b_unity():
|
||||
# test limiting case that betabinom(n, 1, 1) is a discrete uniform
|
||||
# distribution from 0 to n
|
||||
n = 20
|
||||
k = np.arange(n + 1)
|
||||
p = betabinom(n, 1, 1).pmf(k)
|
||||
expected = np.repeat(1 / (n + 1), n + 1)
|
||||
assert_almost_equal(p, expected)
|
||||
|
||||
|
||||
def test_betabinom_bernoulli():
|
||||
# test limiting case that betabinom(1, a, b) = bernoulli(a / (a + b))
|
||||
a = 2.3
|
||||
b = 0.63
|
||||
k = np.arange(2)
|
||||
p = betabinom(1, a, b).pmf(k)
|
||||
expected = bernoulli(a / (a + b)).pmf(k)
|
||||
assert_almost_equal(p, expected)
|
||||
|
||||
|
||||
def test_issue_10317():
|
||||
alpha, n, p = 0.9, 10, 1
|
||||
assert_equal(nbinom.interval(alpha=alpha, n=n, p=p), (0, 0))
|
||||
|
||||
|
||||
def test_issue_11134():
|
||||
alpha, n, p = 0.95, 10, 0
|
||||
assert_equal(binom.interval(alpha=alpha, n=n, p=p), (0, 0))
|
||||
|
||||
|
||||
def test_issue_7406():
|
||||
np.random.seed(0)
|
||||
assert_equal(binom.ppf(np.random.rand(10), 0, 0.5), 0)
|
||||
|
||||
# Also check that endpoints (q=0, q=1) are correct
|
||||
assert_equal(binom.ppf(0, 0, 0.5), -1)
|
||||
assert_equal(binom.ppf(1, 0, 0.5), 0)
|
||||
|
||||
|
||||
def test_issue_5122():
|
||||
p = 0
|
||||
n = np.random.randint(100, size=10)
|
||||
|
||||
x = 0
|
||||
ppf = binom.ppf(x, n, p)
|
||||
assert_equal(ppf, -1)
|
||||
|
||||
x = np.linspace(0.01, 0.99, 10)
|
||||
ppf = binom.ppf(x, n, p)
|
||||
assert_equal(ppf, 0)
|
||||
|
||||
x = 1
|
||||
ppf = binom.ppf(x, n, p)
|
||||
assert_equal(ppf, n)
|
||||
|
||||
|
||||
def test_issue_1603():
|
||||
assert_equal(binom(1000, np.logspace(-3, -100)).ppf(0.01), 0)
|
||||
|
||||
|
||||
def test_issue_5503():
|
||||
p = 0.5
|
||||
x = np.logspace(3, 14, 12)
|
||||
assert_allclose(binom.cdf(x, 2*x, p), 0.5, atol=1e-2)
|
||||
|
||||
|
||||
@pytest.mark.parametrize('x, n, p, cdf_desired', [
|
||||
(300, 1000, 3/10, 0.51559351981411995636),
|
||||
(3000, 10000, 3/10, 0.50493298381929698016),
|
||||
(30000, 100000, 3/10, 0.50156000591726422864),
|
||||
(300000, 1000000, 3/10, 0.50049331906666960038),
|
||||
(3000000, 10000000, 3/10, 0.50015600124585261196),
|
||||
(30000000, 100000000, 3/10, 0.50004933192735230102),
|
||||
(30010000, 100000000, 3/10, 0.98545384016570790717),
|
||||
(29990000, 100000000, 3/10, 0.01455017177985268670),
|
||||
(29950000, 100000000, 3/10, 5.02250963487432024943e-28),
|
||||
])
|
||||
def test_issue_5503pt2(x, n, p, cdf_desired):
|
||||
assert_allclose(binom.cdf(x, n, p), cdf_desired)
|
||||
|
||||
|
||||
def test_issue_5503pt3():
|
||||
# From Wolfram Alpha: CDF[BinomialDistribution[1e12, 1e-12], 2]
|
||||
assert_allclose(binom.cdf(2, 10**12, 10**-12), 0.91969860292869777384)
|
||||
|
||||
|
||||
def test_issue_6682():
|
||||
# Reference value from R:
|
||||
# options(digits=16)
|
||||
# print(pnbinom(250, 50, 32/63, lower.tail=FALSE))
|
||||
assert_allclose(nbinom.sf(250, 50, 32./63.), 1.460458510976452e-35)
|
||||
|
||||
|
||||
def test_skellam_gh11474():
|
||||
# test issue reported in gh-11474 caused by `cdfchn`
|
||||
mu = [1, 10, 100, 1000, 5000, 5050, 5100, 5250, 6000]
|
||||
cdf = skellam.cdf(0, mu, mu)
|
||||
# generated in R
|
||||
# library(skellam)
|
||||
# options(digits = 16)
|
||||
# mu = c(1, 10, 100, 1000, 5000, 5050, 5100, 5250, 6000)
|
||||
# pskellam(0, mu, mu, TRUE)
|
||||
cdf_expected = [0.6542541612768356, 0.5448901559424127, 0.5141135799745580,
|
||||
0.5044605891382528, 0.5019947363350450, 0.5019848365953181,
|
||||
0.5019750827993392, 0.5019466621805060, 0.5018209330219539]
|
||||
assert_allclose(cdf, cdf_expected)
|
||||
|
||||
|
||||
class TestZipfian:
|
||||
def test_zipfian_asymptotic(self):
|
||||
# test limiting case that zipfian(a, n) -> zipf(a) as n-> oo
|
||||
a = 6.5
|
||||
N = 10000000
|
||||
k = np.arange(1, 21)
|
||||
assert_allclose(zipfian.pmf(k, a, N), zipf.pmf(k, a))
|
||||
assert_allclose(zipfian.cdf(k, a, N), zipf.cdf(k, a))
|
||||
assert_allclose(zipfian.sf(k, a, N), zipf.sf(k, a))
|
||||
assert_allclose(zipfian.stats(a, N, moments='msvk'),
|
||||
zipf.stats(a, moments='msvk'))
|
||||
|
||||
def test_zipfian_continuity(self):
|
||||
# test that zipfian(0.999999, n) ~ zipfian(1.000001, n)
|
||||
# (a = 1 switches between methods of calculating harmonic sum)
|
||||
alt1, agt1 = 0.99999999, 1.00000001
|
||||
N = 30
|
||||
k = np.arange(1, N + 1)
|
||||
assert_allclose(zipfian.pmf(k, alt1, N), zipfian.pmf(k, agt1, N),
|
||||
rtol=5e-7)
|
||||
assert_allclose(zipfian.cdf(k, alt1, N), zipfian.cdf(k, agt1, N),
|
||||
rtol=5e-7)
|
||||
assert_allclose(zipfian.sf(k, alt1, N), zipfian.sf(k, agt1, N),
|
||||
rtol=5e-7)
|
||||
assert_allclose(zipfian.stats(alt1, N, moments='msvk'),
|
||||
zipfian.stats(agt1, N, moments='msvk'), rtol=5e-7)
|
||||
|
||||
def test_zipfian_R(self):
|
||||
# test against R VGAM package
|
||||
# library(VGAM)
|
||||
# k <- c(13, 16, 1, 4, 4, 8, 10, 19, 5, 7)
|
||||
# a <- c(1.56712977, 3.72656295, 5.77665117, 9.12168729, 5.79977172,
|
||||
# 4.92784796, 9.36078764, 4.3739616 , 7.48171872, 4.6824154)
|
||||
# n <- c(70, 80, 48, 65, 83, 89, 50, 30, 20, 20)
|
||||
# pmf <- dzipf(k, N = n, shape = a)
|
||||
# cdf <- pzipf(k, N = n, shape = a)
|
||||
# print(pmf)
|
||||
# print(cdf)
|
||||
np.random.seed(0)
|
||||
k = np.random.randint(1, 20, size=10)
|
||||
a = np.random.rand(10)*10 + 1
|
||||
n = np.random.randint(1, 100, size=10)
|
||||
pmf = [8.076972e-03, 2.950214e-05, 9.799333e-01, 3.216601e-06,
|
||||
3.158895e-04, 3.412497e-05, 4.350472e-10, 2.405773e-06,
|
||||
5.860662e-06, 1.053948e-04]
|
||||
cdf = [0.8964133, 0.9998666, 0.9799333, 0.9999995, 0.9998584,
|
||||
0.9999458, 1.0000000, 0.9999920, 0.9999977, 0.9998498]
|
||||
# skip the first point; zipUC is not accurate for low a, n
|
||||
assert_allclose(zipfian.pmf(k, a, n)[1:], pmf[1:], rtol=1e-6)
|
||||
assert_allclose(zipfian.cdf(k, a, n)[1:], cdf[1:], rtol=5e-5)
|
||||
|
||||
np.random.seed(0)
|
||||
naive_tests = np.vstack((np.logspace(-2, 1, 10),
|
||||
np.random.randint(2, 40, 10))).T
|
||||
|
||||
@pytest.mark.parametrize("a, n", naive_tests)
|
||||
def test_zipfian_naive(self, a, n):
|
||||
# test against bare-bones implementation
|
||||
|
||||
@np.vectorize
|
||||
def Hns(n, s):
|
||||
"""Naive implementation of harmonic sum"""
|
||||
return (1/np.arange(1, n+1)**s).sum()
|
||||
|
||||
@np.vectorize
|
||||
def pzip(k, a, n):
|
||||
"""Naive implementation of zipfian pmf"""
|
||||
if k < 1 or k > n:
|
||||
return 0.
|
||||
else:
|
||||
return 1 / k**a / Hns(n, a)
|
||||
|
||||
k = np.arange(n+1)
|
||||
pmf = pzip(k, a, n)
|
||||
cdf = np.cumsum(pmf)
|
||||
mean = np.average(k, weights=pmf)
|
||||
var = np.average((k - mean)**2, weights=pmf)
|
||||
std = var**0.5
|
||||
skew = np.average(((k-mean)/std)**3, weights=pmf)
|
||||
kurtosis = np.average(((k-mean)/std)**4, weights=pmf) - 3
|
||||
assert_allclose(zipfian.pmf(k, a, n), pmf)
|
||||
assert_allclose(zipfian.cdf(k, a, n), cdf)
|
||||
assert_allclose(zipfian.stats(a, n, moments="mvsk"),
|
||||
[mean, var, skew, kurtosis])
|
||||
|
||||
|
||||
class TestNCH():
|
||||
np.random.seed(2) # seeds 0 and 1 had some xl = xu; randint failed
|
||||
shape = (2, 4, 3)
|
||||
max_m = 100
|
||||
m1 = np.random.randint(1, max_m, size=shape) # red balls
|
||||
m2 = np.random.randint(1, max_m, size=shape) # white balls
|
||||
N = m1 + m2 # total balls
|
||||
n = randint.rvs(0, N, size=N.shape) # number of draws
|
||||
xl = np.maximum(0, n-m2) # lower bound of support
|
||||
xu = np.minimum(n, m1) # upper bound of support
|
||||
x = randint.rvs(xl, xu, size=xl.shape)
|
||||
odds = np.random.rand(*x.shape)*2
|
||||
|
||||
# test output is more readable when function names (strings) are passed
|
||||
@pytest.mark.parametrize('dist_name',
|
||||
['nchypergeom_fisher', 'nchypergeom_wallenius'])
|
||||
def test_nch_hypergeom(self, dist_name):
|
||||
# Both noncentral hypergeometric distributions reduce to the
|
||||
# hypergeometric distribution when odds = 1
|
||||
dists = {'nchypergeom_fisher': nchypergeom_fisher,
|
||||
'nchypergeom_wallenius': nchypergeom_wallenius}
|
||||
dist = dists[dist_name]
|
||||
x, N, m1, n = self.x, self.N, self.m1, self.n
|
||||
assert_allclose(dist.pmf(x, N, m1, n, odds=1),
|
||||
hypergeom.pmf(x, N, m1, n))
|
||||
|
||||
def test_nchypergeom_fisher_naive(self):
|
||||
# test against a very simple implementation
|
||||
x, N, m1, n, odds = self.x, self.N, self.m1, self.n, self.odds
|
||||
|
||||
@np.vectorize
|
||||
def pmf_mean_var(x, N, m1, n, w):
|
||||
# simple implementation of nchypergeom_fisher pmf
|
||||
m2 = N - m1
|
||||
xl = np.maximum(0, n-m2)
|
||||
xu = np.minimum(n, m1)
|
||||
|
||||
def f(x):
|
||||
t1 = special_binom(m1, x)
|
||||
t2 = special_binom(m2, n - x)
|
||||
return t1 * t2 * w**x
|
||||
|
||||
def P(k):
|
||||
return sum((f(y)*y**k for y in range(xl, xu + 1)))
|
||||
|
||||
P0 = P(0)
|
||||
P1 = P(1)
|
||||
P2 = P(2)
|
||||
pmf = f(x) / P0
|
||||
mean = P1 / P0
|
||||
var = P2 / P0 - (P1 / P0)**2
|
||||
return pmf, mean, var
|
||||
|
||||
pmf, mean, var = pmf_mean_var(x, N, m1, n, odds)
|
||||
assert_allclose(nchypergeom_fisher.pmf(x, N, m1, n, odds), pmf)
|
||||
assert_allclose(nchypergeom_fisher.stats(N, m1, n, odds, moments='m'),
|
||||
mean)
|
||||
assert_allclose(nchypergeom_fisher.stats(N, m1, n, odds, moments='v'),
|
||||
var)
|
||||
|
||||
def test_nchypergeom_wallenius_naive(self):
|
||||
# test against a very simple implementation
|
||||
|
||||
np.random.seed(2)
|
||||
shape = (2, 4, 3)
|
||||
max_m = 100
|
||||
m1 = np.random.randint(1, max_m, size=shape)
|
||||
m2 = np.random.randint(1, max_m, size=shape)
|
||||
N = m1 + m2
|
||||
n = randint.rvs(0, N, size=N.shape)
|
||||
xl = np.maximum(0, n-m2)
|
||||
xu = np.minimum(n, m1)
|
||||
x = randint.rvs(xl, xu, size=xl.shape)
|
||||
w = np.random.rand(*x.shape)*2
|
||||
|
||||
def support(N, m1, n, w):
|
||||
m2 = N - m1
|
||||
xl = np.maximum(0, n-m2)
|
||||
xu = np.minimum(n, m1)
|
||||
return xl, xu
|
||||
|
||||
@np.vectorize
|
||||
def mean(N, m1, n, w):
|
||||
m2 = N - m1
|
||||
xl, xu = support(N, m1, n, w)
|
||||
|
||||
def fun(u):
|
||||
return u/m1 + (1 - (n-u)/m2)**w - 1
|
||||
|
||||
return root_scalar(fun, bracket=(xl, xu)).root
|
||||
|
||||
with suppress_warnings() as sup:
|
||||
sup.filter(RuntimeWarning,
|
||||
message="invalid value encountered in mean")
|
||||
assert_allclose(nchypergeom_wallenius.mean(N, m1, n, w),
|
||||
mean(N, m1, n, w), rtol=2e-2)
|
||||
|
||||
@np.vectorize
|
||||
def variance(N, m1, n, w):
|
||||
m2 = N - m1
|
||||
u = mean(N, m1, n, w)
|
||||
a = u * (m1 - u)
|
||||
b = (n-u)*(u + m2 - n)
|
||||
return N*a*b / ((N-1) * (m1*b + m2*a))
|
||||
|
||||
with suppress_warnings() as sup:
|
||||
sup.filter(RuntimeWarning,
|
||||
message="invalid value encountered in mean")
|
||||
assert_allclose(
|
||||
nchypergeom_wallenius.stats(N, m1, n, w, moments='v'),
|
||||
variance(N, m1, n, w),
|
||||
rtol=5e-2
|
||||
)
|
||||
|
||||
@np.vectorize
|
||||
def pmf(x, N, m1, n, w):
|
||||
m2 = N - m1
|
||||
xl, xu = support(N, m1, n, w)
|
||||
|
||||
def integrand(t):
|
||||
D = w*(m1 - x) + (m2 - (n-x))
|
||||
res = (1-t**(w/D))**x * (1-t**(1/D))**(n-x)
|
||||
return res
|
||||
|
||||
def f(x):
|
||||
t1 = special_binom(m1, x)
|
||||
t2 = special_binom(m2, n - x)
|
||||
the_integral = quad(integrand, 0, 1,
|
||||
epsrel=1e-16, epsabs=1e-16)
|
||||
return t1 * t2 * the_integral[0]
|
||||
|
||||
return f(x)
|
||||
|
||||
pmf0 = pmf(x, N, m1, n, w)
|
||||
pmf1 = nchypergeom_wallenius.pmf(x, N, m1, n, w)
|
||||
|
||||
atol, rtol = 1e-6, 1e-6
|
||||
i = np.abs(pmf1 - pmf0) < atol + rtol*np.abs(pmf0)
|
||||
assert(i.sum() > np.prod(shape) / 2) # works at least half the time
|
||||
|
||||
# for those that fail, discredit the naive implementation
|
||||
for N, m1, n, w in zip(N[~i], m1[~i], n[~i], w[~i]):
|
||||
# get the support
|
||||
m2 = N - m1
|
||||
xl, xu = support(N, m1, n, w)
|
||||
x = np.arange(xl, xu + 1)
|
||||
|
||||
# calculate sum of pmf over the support
|
||||
# the naive implementation is very wrong in these cases
|
||||
assert pmf(x, N, m1, n, w).sum() < .5
|
||||
assert_allclose(nchypergeom_wallenius.pmf(x, N, m1, n, w).sum(), 1)
|
||||
|
||||
def test_wallenius_against_mpmath(self):
|
||||
# precompute data with mpmath since naive implementation above
|
||||
# is not reliable. See source code in gh-13330.
|
||||
M = 50
|
||||
n = 30
|
||||
N = 20
|
||||
odds = 2.25
|
||||
# Expected results, computed with mpmath.
|
||||
sup = np.arange(21)
|
||||
pmf = np.array([3.699003068656875e-20,
|
||||
5.89398584245431e-17,
|
||||
2.1594437742911123e-14,
|
||||
3.221458044649955e-12,
|
||||
2.4658279241205077e-10,
|
||||
1.0965862603981212e-08,
|
||||
3.057890479665704e-07,
|
||||
5.622818831643761e-06,
|
||||
7.056482841531681e-05,
|
||||
0.000618899425358671,
|
||||
0.003854172932571669,
|
||||
0.01720592676256026,
|
||||
0.05528844897093792,
|
||||
0.12772363313574242,
|
||||
0.21065898367825722,
|
||||
0.24465958845359234,
|
||||
0.1955114898110033,
|
||||
0.10355390084949237,
|
||||
0.03414490375225675,
|
||||
0.006231989845775931,
|
||||
0.0004715577304677075])
|
||||
mean = 14.808018384813426
|
||||
var = 2.6085975877923717
|
||||
|
||||
# nchypergeom_wallenius.pmf returns 0 for pmf(0) and pmf(1), and pmf(2)
|
||||
# has only three digits of accuracy (~ 2.1511e-14).
|
||||
assert_allclose(nchypergeom_wallenius.pmf(sup, M, n, N, odds), pmf,
|
||||
rtol=1e-13, atol=1e-13)
|
||||
assert_allclose(nchypergeom_wallenius.mean(M, n, N, odds),
|
||||
mean, rtol=1e-13)
|
||||
assert_allclose(nchypergeom_wallenius.var(M, n, N, odds),
|
||||
var, rtol=1e-11)
|
||||
|
||||
@pytest.mark.parametrize('dist_name',
|
||||
['nchypergeom_fisher', 'nchypergeom_wallenius'])
|
||||
def test_rvs_shape(self, dist_name):
|
||||
# Check that when given a size with more dimensions than the
|
||||
# dimensions of the broadcast parameters, rvs returns an array
|
||||
# with the correct shape.
|
||||
dists = {'nchypergeom_fisher': nchypergeom_fisher,
|
||||
'nchypergeom_wallenius': nchypergeom_wallenius}
|
||||
dist = dists[dist_name]
|
||||
x = dist.rvs(50, 30, [[10], [20]], [0.5, 1.0, 2.0], size=(5, 1, 2, 3))
|
||||
assert x.shape == (5, 1, 2, 3)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("mu, q, expected",
|
||||
[[10, 120, -1.240089881791596e-38],
|
||||
[1500, 0, -86.61466680572661]])
|
||||
def test_nbinom_11465(mu, q, expected):
|
||||
# test nbinom.logcdf at extreme tails
|
||||
size = 20
|
||||
n, p = size, size/(size+mu)
|
||||
# In R:
|
||||
# options(digits=16)
|
||||
# pnbinom(mu=10, size=20, q=120, log.p=TRUE)
|
||||
assert_allclose(nbinom.logcdf(q, n, p), expected)
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,287 @@
|
||||
|
||||
import numpy as np
|
||||
from numpy.testing import assert_equal, assert_allclose
|
||||
# avoid new uses of the following; prefer assert/np.testing.assert_allclose
|
||||
from numpy.testing import (assert_, assert_almost_equal,
|
||||
assert_array_almost_equal)
|
||||
|
||||
import pytest
|
||||
from pytest import raises as assert_raises
|
||||
import scipy.stats as stats
|
||||
|
||||
|
||||
class TestEntropy:
|
||||
def test_entropy_positive(self):
|
||||
# See ticket #497
|
||||
pk = [0.5, 0.2, 0.3]
|
||||
qk = [0.1, 0.25, 0.65]
|
||||
eself = stats.entropy(pk, pk)
|
||||
edouble = stats.entropy(pk, qk)
|
||||
assert_(0.0 == eself)
|
||||
assert_(edouble >= 0.0)
|
||||
|
||||
def test_entropy_base(self):
|
||||
pk = np.ones(16, float)
|
||||
S = stats.entropy(pk, base=2.)
|
||||
assert_(abs(S - 4.) < 1.e-5)
|
||||
|
||||
qk = np.ones(16, float)
|
||||
qk[:8] = 2.
|
||||
S = stats.entropy(pk, qk)
|
||||
S2 = stats.entropy(pk, qk, base=2.)
|
||||
assert_(abs(S/S2 - np.log(2.)) < 1.e-5)
|
||||
|
||||
def test_entropy_zero(self):
|
||||
# Test for PR-479
|
||||
assert_almost_equal(stats.entropy([0, 1, 2]), 0.63651416829481278,
|
||||
decimal=12)
|
||||
|
||||
def test_entropy_2d(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
qk = [[0.2, 0.1], [0.3, 0.6], [0.5, 0.3]]
|
||||
assert_array_almost_equal(stats.entropy(pk, qk),
|
||||
[0.1933259, 0.18609809])
|
||||
|
||||
def test_entropy_2d_zero(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
qk = [[0.0, 0.1], [0.3, 0.6], [0.5, 0.3]]
|
||||
assert_array_almost_equal(stats.entropy(pk, qk),
|
||||
[np.inf, 0.18609809])
|
||||
|
||||
pk[0][0] = 0.0
|
||||
assert_array_almost_equal(stats.entropy(pk, qk),
|
||||
[0.17403988, 0.18609809])
|
||||
|
||||
def test_entropy_base_2d_nondefault_axis(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
assert_array_almost_equal(stats.entropy(pk, axis=1),
|
||||
[0.63651417, 0.63651417, 0.66156324])
|
||||
|
||||
def test_entropy_2d_nondefault_axis(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
qk = [[0.2, 0.1], [0.3, 0.6], [0.5, 0.3]]
|
||||
assert_array_almost_equal(stats.entropy(pk, qk, axis=1),
|
||||
[0.231049, 0.231049, 0.127706])
|
||||
|
||||
def test_entropy_raises_value_error(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
qk = [[0.1, 0.2], [0.6, 0.3]]
|
||||
assert_raises(ValueError, stats.entropy, pk, qk)
|
||||
|
||||
def test_base_entropy_with_axis_0_is_equal_to_default(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
assert_array_almost_equal(stats.entropy(pk, axis=0),
|
||||
stats.entropy(pk))
|
||||
|
||||
def test_entropy_with_axis_0_is_equal_to_default(self):
|
||||
pk = [[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]]
|
||||
qk = [[0.2, 0.1], [0.3, 0.6], [0.5, 0.3]]
|
||||
assert_array_almost_equal(stats.entropy(pk, qk, axis=0),
|
||||
stats.entropy(pk, qk))
|
||||
|
||||
def test_base_entropy_transposed(self):
|
||||
pk = np.array([[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]])
|
||||
assert_array_almost_equal(stats.entropy(pk.T).T,
|
||||
stats.entropy(pk, axis=1))
|
||||
|
||||
def test_entropy_transposed(self):
|
||||
pk = np.array([[0.1, 0.2], [0.6, 0.3], [0.3, 0.5]])
|
||||
qk = np.array([[0.2, 0.1], [0.3, 0.6], [0.5, 0.3]])
|
||||
assert_array_almost_equal(stats.entropy(pk.T, qk.T).T,
|
||||
stats.entropy(pk, qk, axis=1))
|
||||
|
||||
def test_entropy_broadcasting(self):
|
||||
np.random.rand(0)
|
||||
x = np.random.rand(3)
|
||||
y = np.random.rand(2, 1)
|
||||
res = stats.entropy(x, y, axis=-1)
|
||||
assert_equal(res[0], stats.entropy(x, y[0]))
|
||||
assert_equal(res[1], stats.entropy(x, y[1]))
|
||||
|
||||
def test_entropy_shape_mismatch(self):
|
||||
x = np.random.rand(10, 1, 12)
|
||||
y = np.random.rand(11, 2)
|
||||
message = "shape mismatch: objects cannot be broadcast"
|
||||
with pytest.raises(ValueError, match=message):
|
||||
stats.entropy(x, y)
|
||||
|
||||
def test_input_validation(self):
|
||||
x = np.random.rand(10)
|
||||
message = "`base` must be a positive number."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
stats.entropy(x, base=-2)
|
||||
|
||||
|
||||
class TestDifferentialEntropy:
|
||||
"""
|
||||
Vasicek results are compared with the R package vsgoftest.
|
||||
|
||||
# library(vsgoftest)
|
||||
#
|
||||
# samp <- c(<values>)
|
||||
# entropy.estimate(x = samp, window = <window_length>)
|
||||
|
||||
"""
|
||||
|
||||
def test_differential_entropy_vasicek(self):
|
||||
|
||||
random_state = np.random.RandomState(0)
|
||||
values = random_state.standard_normal(100)
|
||||
|
||||
entropy = stats.differential_entropy(values, method='vasicek')
|
||||
assert_allclose(entropy, 1.342551, rtol=1e-6)
|
||||
|
||||
entropy = stats.differential_entropy(values, window_length=1,
|
||||
method='vasicek')
|
||||
assert_allclose(entropy, 1.122044, rtol=1e-6)
|
||||
|
||||
entropy = stats.differential_entropy(values, window_length=8,
|
||||
method='vasicek')
|
||||
assert_allclose(entropy, 1.349401, rtol=1e-6)
|
||||
|
||||
def test_differential_entropy_vasicek_2d_nondefault_axis(self):
|
||||
random_state = np.random.RandomState(0)
|
||||
values = random_state.standard_normal((3, 100))
|
||||
|
||||
entropy = stats.differential_entropy(values, axis=1, method='vasicek')
|
||||
assert_allclose(
|
||||
entropy,
|
||||
[1.342551, 1.341826, 1.293775],
|
||||
rtol=1e-6,
|
||||
)
|
||||
|
||||
entropy = stats.differential_entropy(values, axis=1, window_length=1,
|
||||
method='vasicek')
|
||||
assert_allclose(
|
||||
entropy,
|
||||
[1.122044, 1.102944, 1.129616],
|
||||
rtol=1e-6,
|
||||
)
|
||||
|
||||
entropy = stats.differential_entropy(values, axis=1, window_length=8,
|
||||
method='vasicek')
|
||||
assert_allclose(
|
||||
entropy,
|
||||
[1.349401, 1.338514, 1.292332],
|
||||
rtol=1e-6,
|
||||
)
|
||||
|
||||
def test_differential_entropy_raises_value_error(self):
|
||||
random_state = np.random.RandomState(0)
|
||||
values = random_state.standard_normal((3, 100))
|
||||
|
||||
error_str = (
|
||||
r"Window length \({window_length}\) must be positive and less "
|
||||
r"than half the sample size \({sample_size}\)."
|
||||
)
|
||||
|
||||
sample_size = values.shape[1]
|
||||
|
||||
for window_length in {-1, 0, sample_size//2, sample_size}:
|
||||
|
||||
formatted_error_str = error_str.format(
|
||||
window_length=window_length,
|
||||
sample_size=sample_size,
|
||||
)
|
||||
|
||||
with assert_raises(ValueError, match=formatted_error_str):
|
||||
stats.differential_entropy(
|
||||
values,
|
||||
window_length=window_length,
|
||||
axis=1,
|
||||
)
|
||||
|
||||
def test_base_differential_entropy_with_axis_0_is_equal_to_default(self):
|
||||
random_state = np.random.RandomState(0)
|
||||
values = random_state.standard_normal((100, 3))
|
||||
|
||||
entropy = stats.differential_entropy(values, axis=0)
|
||||
default_entropy = stats.differential_entropy(values)
|
||||
assert_allclose(entropy, default_entropy)
|
||||
|
||||
def test_base_differential_entropy_transposed(self):
|
||||
random_state = np.random.RandomState(0)
|
||||
values = random_state.standard_normal((3, 100))
|
||||
|
||||
assert_allclose(
|
||||
stats.differential_entropy(values.T).T,
|
||||
stats.differential_entropy(values, axis=1),
|
||||
)
|
||||
|
||||
def test_input_validation(self):
|
||||
x = np.random.rand(10)
|
||||
|
||||
message = "`base` must be a positive number or `None`."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
stats.differential_entropy(x, base=-2)
|
||||
|
||||
message = "`method` must be one of..."
|
||||
with pytest.raises(ValueError, match=message):
|
||||
stats.differential_entropy(x, method='ekki-ekki')
|
||||
|
||||
@pytest.mark.parametrize('method', ['vasicek', 'van es',
|
||||
'ebrahimi', 'correa'])
|
||||
def test_consistency(self, method):
|
||||
# test that method is a consistent estimator
|
||||
n = 10000 if method == 'correa' else 1000000
|
||||
rvs = stats.norm.rvs(size=n, random_state=0)
|
||||
expected = stats.norm.entropy()
|
||||
res = stats.differential_entropy(rvs, method=method)
|
||||
assert_allclose(res, expected, rtol=0.005)
|
||||
|
||||
# values from differential_entropy reference [6], table 1, n=50, m=7
|
||||
norm_rmse_std_cases = { # method: (RMSE, STD)
|
||||
'vasicek': (0.198, 0.109),
|
||||
'van es': (0.212, 0.110),
|
||||
'correa': (0.135, 0.112),
|
||||
'ebrahimi': (0.128, 0.109)
|
||||
}
|
||||
|
||||
@pytest.mark.parametrize('method, expected',
|
||||
list(norm_rmse_std_cases.items()))
|
||||
def test_norm_rmse_std(self, method, expected):
|
||||
# test that RMSE and standard deviation of estimators matches values
|
||||
# given in differential_entropy reference [6]. Incidentally, also
|
||||
# tests vectorization.
|
||||
reps, n, m = 10000, 50, 7
|
||||
rmse_expected, std_expected = expected
|
||||
rvs = stats.norm.rvs(size=(reps, n), random_state=0)
|
||||
true_entropy = stats.norm.entropy()
|
||||
res = stats.differential_entropy(rvs, window_length=m,
|
||||
method=method, axis=-1)
|
||||
assert_allclose(np.sqrt(np.mean((res - true_entropy)**2)),
|
||||
rmse_expected, atol=0.005)
|
||||
assert_allclose(np.std(res), std_expected, atol=0.002)
|
||||
|
||||
# values from differential_entropy reference [6], table 2, n=50, m=7
|
||||
expon_rmse_std_cases = { # method: (RMSE, STD)
|
||||
'vasicek': (0.194, 0.148),
|
||||
'van es': (0.179, 0.149),
|
||||
'correa': (0.155, 0.152),
|
||||
'ebrahimi': (0.151, 0.148)
|
||||
}
|
||||
|
||||
@pytest.mark.parametrize('method, expected',
|
||||
list(expon_rmse_std_cases.items()))
|
||||
def test_expon_rmse_std(self, method, expected):
|
||||
# test that RMSE and standard deviation of estimators matches values
|
||||
# given in differential_entropy reference [6]. Incidentally, also
|
||||
# tests vectorization.
|
||||
reps, n, m = 10000, 50, 7
|
||||
rmse_expected, std_expected = expected
|
||||
rvs = stats.expon.rvs(size=(reps, n), random_state=0)
|
||||
true_entropy = stats.expon.entropy()
|
||||
res = stats.differential_entropy(rvs, window_length=m,
|
||||
method=method, axis=-1)
|
||||
assert_allclose(np.sqrt(np.mean((res - true_entropy)**2)),
|
||||
rmse_expected, atol=0.005)
|
||||
assert_allclose(np.std(res), std_expected, atol=0.002)
|
||||
|
||||
@pytest.mark.parametrize('n, method', [(8, 'van es'),
|
||||
(12, 'ebrahimi'),
|
||||
(1001, 'vasicek')])
|
||||
def test_method_auto(self, n, method):
|
||||
rvs = stats.norm.rvs(size=(n,), random_state=0)
|
||||
res1 = stats.differential_entropy(rvs)
|
||||
res2 = stats.differential_entropy(rvs, method=method)
|
||||
assert res1 == res2
|
||||
@@ -0,0 +1,141 @@
|
||||
import os
|
||||
|
||||
import numpy as np
|
||||
from numpy.testing import assert_allclose
|
||||
import pytest
|
||||
from scipy import stats
|
||||
|
||||
from .test_continuous_basic import distcont
|
||||
|
||||
# this is not a proper statistical test for convergence, but only
|
||||
# verifies that the estimate and true values don't differ by too much
|
||||
|
||||
fit_sizes = [1000, 5000, 10000] # sample sizes to try
|
||||
|
||||
thresh_percent = 0.25 # percent of true parameters for fail cut-off
|
||||
thresh_min = 0.75 # minimum difference estimate - true to fail test
|
||||
|
||||
mle_failing_fits = [
|
||||
'burr',
|
||||
'chi2',
|
||||
'gausshyper',
|
||||
'genexpon',
|
||||
'gengamma',
|
||||
'kappa4',
|
||||
'ksone',
|
||||
'kstwo',
|
||||
'mielke',
|
||||
'ncf',
|
||||
'ncx2',
|
||||
'pearson3',
|
||||
'powerlognorm',
|
||||
'truncexpon',
|
||||
'tukeylambda',
|
||||
'vonmises',
|
||||
'levy_stable',
|
||||
'trapezoid',
|
||||
'studentized_range'
|
||||
]
|
||||
|
||||
mm_failing_fits = ['alpha', 'betaprime', 'burr', 'burr12', 'cauchy', 'chi',
|
||||
'chi2', 'crystalball', 'dgamma', 'dweibull', 'f',
|
||||
'fatiguelife', 'fisk', 'foldcauchy', 'genextreme',
|
||||
'gengamma', 'genhyperbolic', 'gennorm', 'genpareto',
|
||||
'halfcauchy', 'invgamma', 'invweibull', 'johnsonsu',
|
||||
'kappa3', 'ksone', 'kstwo', 'levy', 'levy_l',
|
||||
'levy_stable', 'loglaplace', 'lomax', 'mielke', 'nakagami',
|
||||
'ncf', 'nct', 'ncx2', 'pareto', 'powerlognorm', 'powernorm',
|
||||
'skewcauchy', 't',
|
||||
'trapezoid', 'triang', 'tukeylambda', 'studentized_range']
|
||||
|
||||
# not sure if these fail, but they caused my patience to fail
|
||||
mm_slow_fits = ['argus', 'exponpow', 'exponweib', 'gausshyper', 'genexpon',
|
||||
'genhalflogistic', 'halfgennorm', 'gompertz', 'johnsonsb',
|
||||
'kappa4', 'kstwobign', 'recipinvgauss', 'skewnorm',
|
||||
'truncexpon', 'vonmises', 'vonmises_line']
|
||||
|
||||
failing_fits = {"MM": mm_failing_fits + mm_slow_fits, "MLE": mle_failing_fits}
|
||||
|
||||
# Don't run the fit test on these:
|
||||
skip_fit = [
|
||||
'erlang', # Subclass of gamma, generates a warning.
|
||||
]
|
||||
|
||||
|
||||
def cases_test_cont_fit():
|
||||
# this tests the closeness of the estimated parameters to the true
|
||||
# parameters with fit method of continuous distributions
|
||||
# Note: is slow, some distributions don't converge with sample
|
||||
# size <= 10000
|
||||
for distname, arg in distcont:
|
||||
if distname not in skip_fit:
|
||||
yield distname, arg
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
@pytest.mark.parametrize('distname,arg', cases_test_cont_fit())
|
||||
@pytest.mark.parametrize('method', ["MLE", 'MM'])
|
||||
def test_cont_fit(distname, arg, method):
|
||||
if distname in failing_fits[method]:
|
||||
# Skip failing fits unless overridden
|
||||
try:
|
||||
xfail = not int(os.environ['SCIPY_XFAIL'])
|
||||
except Exception:
|
||||
xfail = True
|
||||
if xfail:
|
||||
msg = "Fitting %s doesn't work reliably yet" % distname
|
||||
msg += (" [Set environment variable SCIPY_XFAIL=1 to run this"
|
||||
" test nevertheless.]")
|
||||
pytest.xfail(msg)
|
||||
|
||||
distfn = getattr(stats, distname)
|
||||
|
||||
truearg = np.hstack([arg, [0.0, 1.0]])
|
||||
diffthreshold = np.max(np.vstack([truearg*thresh_percent,
|
||||
np.full(distfn.numargs+2, thresh_min)]),
|
||||
0)
|
||||
|
||||
for fit_size in fit_sizes:
|
||||
# Note that if a fit succeeds, the other fit_sizes are skipped
|
||||
np.random.seed(1234)
|
||||
|
||||
with np.errstate(all='ignore'):
|
||||
rvs = distfn.rvs(size=fit_size, *arg)
|
||||
est = distfn.fit(rvs, method=method) # start with default values
|
||||
|
||||
diff = est - truearg
|
||||
|
||||
# threshold for location
|
||||
diffthreshold[-2] = np.max([np.abs(rvs.mean())*thresh_percent,
|
||||
thresh_min])
|
||||
|
||||
if np.any(np.isnan(est)):
|
||||
raise AssertionError('nan returned in fit')
|
||||
else:
|
||||
if np.all(np.abs(diff) <= diffthreshold):
|
||||
break
|
||||
else:
|
||||
txt = 'parameter: %s\n' % str(truearg)
|
||||
txt += 'estimated: %s\n' % str(est)
|
||||
txt += 'diff : %s\n' % str(diff)
|
||||
raise AssertionError('fit not very good in %s\n' % distfn.name + txt)
|
||||
|
||||
|
||||
def _check_loc_scale_mle_fit(name, data, desired, atol=None):
|
||||
d = getattr(stats, name)
|
||||
actual = d.fit(data)[-2:]
|
||||
assert_allclose(actual, desired, atol=atol,
|
||||
err_msg='poor mle fit of (loc, scale) in %s' % name)
|
||||
|
||||
|
||||
def test_non_default_loc_scale_mle_fit():
|
||||
data = np.array([1.01, 1.78, 1.78, 1.78, 1.88, 1.88, 1.88, 2.00])
|
||||
_check_loc_scale_mle_fit('uniform', data, [1.01, 0.99], 1e-3)
|
||||
_check_loc_scale_mle_fit('expon', data, [1.01, 0.73875], 1e-3)
|
||||
|
||||
|
||||
def test_expon_fit():
|
||||
"""gh-6167"""
|
||||
data = [0, 0, 0, 0, 2, 2, 2, 2]
|
||||
phat = stats.expon.fit(data, floc=0)
|
||||
assert_allclose(phat, [0, 1.0], atol=1e-3)
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,488 @@
|
||||
from scipy import stats
|
||||
import numpy as np
|
||||
from numpy.testing import (assert_almost_equal, assert_,
|
||||
assert_array_almost_equal, assert_array_almost_equal_nulp, assert_allclose)
|
||||
import pytest
|
||||
from pytest import raises as assert_raises
|
||||
|
||||
|
||||
def test_kde_1d():
|
||||
#some basic tests comparing to normal distribution
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 500
|
||||
xn = np.random.randn(n_basesample)
|
||||
xnmean = xn.mean()
|
||||
xnstd = xn.std(ddof=1)
|
||||
|
||||
# get kde for original sample
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
|
||||
# evaluate the density function for the kde for some points
|
||||
xs = np.linspace(-7,7,501)
|
||||
kdepdf = gkde.evaluate(xs)
|
||||
normpdf = stats.norm.pdf(xs, loc=xnmean, scale=xnstd)
|
||||
intervall = xs[1] - xs[0]
|
||||
|
||||
assert_(np.sum((kdepdf - normpdf)**2)*intervall < 0.01)
|
||||
prob1 = gkde.integrate_box_1d(xnmean, np.inf)
|
||||
prob2 = gkde.integrate_box_1d(-np.inf, xnmean)
|
||||
assert_almost_equal(prob1, 0.5, decimal=1)
|
||||
assert_almost_equal(prob2, 0.5, decimal=1)
|
||||
assert_almost_equal(gkde.integrate_box(xnmean, np.inf), prob1, decimal=13)
|
||||
assert_almost_equal(gkde.integrate_box(-np.inf, xnmean), prob2, decimal=13)
|
||||
|
||||
assert_almost_equal(gkde.integrate_kde(gkde),
|
||||
(kdepdf**2).sum()*intervall, decimal=2)
|
||||
assert_almost_equal(gkde.integrate_gaussian(xnmean, xnstd**2),
|
||||
(kdepdf*normpdf).sum()*intervall, decimal=2)
|
||||
|
||||
|
||||
def test_kde_1d_weighted():
|
||||
#some basic tests comparing to normal distribution
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 500
|
||||
xn = np.random.randn(n_basesample)
|
||||
wn = np.random.rand(n_basesample)
|
||||
xnmean = np.average(xn, weights=wn)
|
||||
xnstd = np.sqrt(np.average((xn-xnmean)**2, weights=wn))
|
||||
|
||||
# get kde for original sample
|
||||
gkde = stats.gaussian_kde(xn, weights=wn)
|
||||
|
||||
# evaluate the density function for the kde for some points
|
||||
xs = np.linspace(-7,7,501)
|
||||
kdepdf = gkde.evaluate(xs)
|
||||
normpdf = stats.norm.pdf(xs, loc=xnmean, scale=xnstd)
|
||||
intervall = xs[1] - xs[0]
|
||||
|
||||
assert_(np.sum((kdepdf - normpdf)**2)*intervall < 0.01)
|
||||
prob1 = gkde.integrate_box_1d(xnmean, np.inf)
|
||||
prob2 = gkde.integrate_box_1d(-np.inf, xnmean)
|
||||
assert_almost_equal(prob1, 0.5, decimal=1)
|
||||
assert_almost_equal(prob2, 0.5, decimal=1)
|
||||
assert_almost_equal(gkde.integrate_box(xnmean, np.inf), prob1, decimal=13)
|
||||
assert_almost_equal(gkde.integrate_box(-np.inf, xnmean), prob2, decimal=13)
|
||||
|
||||
assert_almost_equal(gkde.integrate_kde(gkde),
|
||||
(kdepdf**2).sum()*intervall, decimal=2)
|
||||
assert_almost_equal(gkde.integrate_gaussian(xnmean, xnstd**2),
|
||||
(kdepdf*normpdf).sum()*intervall, decimal=2)
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
def test_kde_2d():
|
||||
#some basic tests comparing to normal distribution
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 500
|
||||
|
||||
mean = np.array([1.0, 3.0])
|
||||
covariance = np.array([[1.0, 2.0], [2.0, 6.0]])
|
||||
|
||||
# Need transpose (shape (2, 500)) for kde
|
||||
xn = np.random.multivariate_normal(mean, covariance, size=n_basesample).T
|
||||
|
||||
# get kde for original sample
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
|
||||
# evaluate the density function for the kde for some points
|
||||
x, y = np.mgrid[-7:7:500j, -7:7:500j]
|
||||
grid_coords = np.vstack([x.ravel(), y.ravel()])
|
||||
kdepdf = gkde.evaluate(grid_coords)
|
||||
kdepdf = kdepdf.reshape(500, 500)
|
||||
|
||||
normpdf = stats.multivariate_normal.pdf(np.dstack([x, y]), mean=mean, cov=covariance)
|
||||
intervall = y.ravel()[1] - y.ravel()[0]
|
||||
|
||||
assert_(np.sum((kdepdf - normpdf)**2) * (intervall**2) < 0.01)
|
||||
|
||||
small = -1e100
|
||||
large = 1e100
|
||||
prob1 = gkde.integrate_box([small, mean[1]], [large, large])
|
||||
prob2 = gkde.integrate_box([small, small], [large, mean[1]])
|
||||
|
||||
assert_almost_equal(prob1, 0.5, decimal=1)
|
||||
assert_almost_equal(prob2, 0.5, decimal=1)
|
||||
assert_almost_equal(gkde.integrate_kde(gkde),
|
||||
(kdepdf**2).sum()*(intervall**2), decimal=2)
|
||||
assert_almost_equal(gkde.integrate_gaussian(mean, covariance),
|
||||
(kdepdf*normpdf).sum()*(intervall**2), decimal=2)
|
||||
|
||||
|
||||
@pytest.mark.slow
|
||||
def test_kde_2d_weighted():
|
||||
#some basic tests comparing to normal distribution
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 500
|
||||
|
||||
mean = np.array([1.0, 3.0])
|
||||
covariance = np.array([[1.0, 2.0], [2.0, 6.0]])
|
||||
|
||||
# Need transpose (shape (2, 500)) for kde
|
||||
xn = np.random.multivariate_normal(mean, covariance, size=n_basesample).T
|
||||
wn = np.random.rand(n_basesample)
|
||||
|
||||
# get kde for original sample
|
||||
gkde = stats.gaussian_kde(xn, weights=wn)
|
||||
|
||||
# evaluate the density function for the kde for some points
|
||||
x, y = np.mgrid[-7:7:500j, -7:7:500j]
|
||||
grid_coords = np.vstack([x.ravel(), y.ravel()])
|
||||
kdepdf = gkde.evaluate(grid_coords)
|
||||
kdepdf = kdepdf.reshape(500, 500)
|
||||
|
||||
normpdf = stats.multivariate_normal.pdf(np.dstack([x, y]), mean=mean, cov=covariance)
|
||||
intervall = y.ravel()[1] - y.ravel()[0]
|
||||
|
||||
assert_(np.sum((kdepdf - normpdf)**2) * (intervall**2) < 0.01)
|
||||
|
||||
small = -1e100
|
||||
large = 1e100
|
||||
prob1 = gkde.integrate_box([small, mean[1]], [large, large])
|
||||
prob2 = gkde.integrate_box([small, small], [large, mean[1]])
|
||||
|
||||
assert_almost_equal(prob1, 0.5, decimal=1)
|
||||
assert_almost_equal(prob2, 0.5, decimal=1)
|
||||
assert_almost_equal(gkde.integrate_kde(gkde),
|
||||
(kdepdf**2).sum()*(intervall**2), decimal=2)
|
||||
assert_almost_equal(gkde.integrate_gaussian(mean, covariance),
|
||||
(kdepdf*normpdf).sum()*(intervall**2), decimal=2)
|
||||
|
||||
|
||||
def test_kde_bandwidth_method():
|
||||
def scotts_factor(kde_obj):
|
||||
"""Same as default, just check that it works."""
|
||||
return np.power(kde_obj.n, -1./(kde_obj.d+4))
|
||||
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 50
|
||||
xn = np.random.randn(n_basesample)
|
||||
|
||||
# Default
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
# Supply a callable
|
||||
gkde2 = stats.gaussian_kde(xn, bw_method=scotts_factor)
|
||||
# Supply a scalar
|
||||
gkde3 = stats.gaussian_kde(xn, bw_method=gkde.factor)
|
||||
|
||||
xs = np.linspace(-7,7,51)
|
||||
kdepdf = gkde.evaluate(xs)
|
||||
kdepdf2 = gkde2.evaluate(xs)
|
||||
assert_almost_equal(kdepdf, kdepdf2)
|
||||
kdepdf3 = gkde3.evaluate(xs)
|
||||
assert_almost_equal(kdepdf, kdepdf3)
|
||||
|
||||
assert_raises(ValueError, stats.gaussian_kde, xn, bw_method='wrongstring')
|
||||
|
||||
|
||||
def test_kde_bandwidth_method_weighted():
|
||||
def scotts_factor(kde_obj):
|
||||
"""Same as default, just check that it works."""
|
||||
return np.power(kde_obj.neff, -1./(kde_obj.d+4))
|
||||
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 50
|
||||
xn = np.random.randn(n_basesample)
|
||||
|
||||
# Default
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
# Supply a callable
|
||||
gkde2 = stats.gaussian_kde(xn, bw_method=scotts_factor)
|
||||
# Supply a scalar
|
||||
gkde3 = stats.gaussian_kde(xn, bw_method=gkde.factor)
|
||||
|
||||
xs = np.linspace(-7,7,51)
|
||||
kdepdf = gkde.evaluate(xs)
|
||||
kdepdf2 = gkde2.evaluate(xs)
|
||||
assert_almost_equal(kdepdf, kdepdf2)
|
||||
kdepdf3 = gkde3.evaluate(xs)
|
||||
assert_almost_equal(kdepdf, kdepdf3)
|
||||
|
||||
assert_raises(ValueError, stats.gaussian_kde, xn, bw_method='wrongstring')
|
||||
|
||||
|
||||
# Subclasses that should stay working (extracted from various sources).
|
||||
# Unfortunately the earlier design of gaussian_kde made it necessary for users
|
||||
# to create these kinds of subclasses, or call _compute_covariance() directly.
|
||||
|
||||
class _kde_subclass1(stats.gaussian_kde):
|
||||
def __init__(self, dataset):
|
||||
self.dataset = np.atleast_2d(dataset)
|
||||
self.d, self.n = self.dataset.shape
|
||||
self.covariance_factor = self.scotts_factor
|
||||
self._compute_covariance()
|
||||
|
||||
|
||||
class _kde_subclass2(stats.gaussian_kde):
|
||||
def __init__(self, dataset):
|
||||
self.covariance_factor = self.scotts_factor
|
||||
super().__init__(dataset)
|
||||
|
||||
|
||||
class _kde_subclass3(stats.gaussian_kde):
|
||||
def __init__(self, dataset, covariance):
|
||||
self.covariance = covariance
|
||||
stats.gaussian_kde.__init__(self, dataset)
|
||||
|
||||
def _compute_covariance(self):
|
||||
self.inv_cov = np.linalg.inv(self.covariance)
|
||||
self._norm_factor = np.sqrt(np.linalg.det(2 * np.pi * self.covariance))
|
||||
|
||||
|
||||
class _kde_subclass4(stats.gaussian_kde):
|
||||
def covariance_factor(self):
|
||||
return 0.5 * self.silverman_factor()
|
||||
|
||||
|
||||
def test_gaussian_kde_subclassing():
|
||||
x1 = np.array([-7, -5, 1, 4, 5], dtype=float)
|
||||
xs = np.linspace(-10, 10, num=50)
|
||||
|
||||
# gaussian_kde itself
|
||||
kde = stats.gaussian_kde(x1)
|
||||
ys = kde(xs)
|
||||
|
||||
# subclass 1
|
||||
kde1 = _kde_subclass1(x1)
|
||||
y1 = kde1(xs)
|
||||
assert_array_almost_equal_nulp(ys, y1, nulp=10)
|
||||
|
||||
# subclass 2
|
||||
kde2 = _kde_subclass2(x1)
|
||||
y2 = kde2(xs)
|
||||
assert_array_almost_equal_nulp(ys, y2, nulp=10)
|
||||
|
||||
# subclass 3
|
||||
kde3 = _kde_subclass3(x1, kde.covariance)
|
||||
y3 = kde3(xs)
|
||||
assert_array_almost_equal_nulp(ys, y3, nulp=10)
|
||||
|
||||
# subclass 4
|
||||
kde4 = _kde_subclass4(x1)
|
||||
y4 = kde4(x1)
|
||||
y_expected = [0.06292987, 0.06346938, 0.05860291, 0.08657652, 0.07904017]
|
||||
|
||||
assert_array_almost_equal(y_expected, y4, decimal=6)
|
||||
|
||||
# Not a subclass, but check for use of _compute_covariance()
|
||||
kde5 = kde
|
||||
kde5.covariance_factor = lambda: kde.factor
|
||||
kde5._compute_covariance()
|
||||
y5 = kde5(xs)
|
||||
assert_array_almost_equal_nulp(ys, y5, nulp=10)
|
||||
|
||||
|
||||
def test_gaussian_kde_covariance_caching():
|
||||
x1 = np.array([-7, -5, 1, 4, 5], dtype=float)
|
||||
xs = np.linspace(-10, 10, num=5)
|
||||
# These expected values are from scipy 0.10, before some changes to
|
||||
# gaussian_kde. They were not compared with any external reference.
|
||||
y_expected = [0.02463386, 0.04689208, 0.05395444, 0.05337754, 0.01664475]
|
||||
|
||||
# Set the bandwidth, then reset it to the default.
|
||||
kde = stats.gaussian_kde(x1)
|
||||
kde.set_bandwidth(bw_method=0.5)
|
||||
kde.set_bandwidth(bw_method='scott')
|
||||
y2 = kde(xs)
|
||||
|
||||
assert_array_almost_equal(y_expected, y2, decimal=7)
|
||||
|
||||
|
||||
def test_gaussian_kde_monkeypatch():
|
||||
"""Ugly, but people may rely on this. See scipy pull request 123,
|
||||
specifically the linked ML thread "Width of the Gaussian in stats.kde".
|
||||
If it is necessary to break this later on, that is to be discussed on ML.
|
||||
"""
|
||||
x1 = np.array([-7, -5, 1, 4, 5], dtype=float)
|
||||
xs = np.linspace(-10, 10, num=50)
|
||||
|
||||
# The old monkeypatched version to get at Silverman's Rule.
|
||||
kde = stats.gaussian_kde(x1)
|
||||
kde.covariance_factor = kde.silverman_factor
|
||||
kde._compute_covariance()
|
||||
y1 = kde(xs)
|
||||
|
||||
# The new saner version.
|
||||
kde2 = stats.gaussian_kde(x1, bw_method='silverman')
|
||||
y2 = kde2(xs)
|
||||
|
||||
assert_array_almost_equal_nulp(y1, y2, nulp=10)
|
||||
|
||||
|
||||
def test_kde_integer_input():
|
||||
"""Regression test for #1181."""
|
||||
x1 = np.arange(5)
|
||||
kde = stats.gaussian_kde(x1)
|
||||
y_expected = [0.13480721, 0.18222869, 0.19514935, 0.18222869, 0.13480721]
|
||||
assert_array_almost_equal(kde(x1), y_expected, decimal=6)
|
||||
|
||||
|
||||
_ftypes = ['float32', 'float64', 'float96', 'float128', 'int32', 'int64']
|
||||
|
||||
@pytest.mark.parametrize("bw_type", _ftypes + ["scott", "silverman"])
|
||||
@pytest.mark.parametrize("weights_type", _ftypes)
|
||||
@pytest.mark.parametrize("dataset_type", _ftypes)
|
||||
@pytest.mark.parametrize("point_type", _ftypes)
|
||||
def test_kde_output_dtype(point_type, dataset_type, weights_type, bw_type):
|
||||
# Check whether the datatypes are available
|
||||
point_type = getattr(np, point_type, None)
|
||||
dataset_type = getattr(np, weights_type, None)
|
||||
weights_type = getattr(np, weights_type, None)
|
||||
|
||||
if bw_type in ["scott", "silverman"]:
|
||||
bw = bw_type
|
||||
else:
|
||||
bw_type = getattr(np, bw_type, None)
|
||||
bw = bw_type(3) if bw_type else None
|
||||
|
||||
if any(dt is None for dt in [point_type, dataset_type, weights_type, bw]):
|
||||
pytest.skip()
|
||||
|
||||
weights = np.arange(5, dtype=weights_type)
|
||||
dataset = np.arange(5, dtype=dataset_type)
|
||||
k = stats.gaussian_kde(dataset, bw_method=bw, weights=weights)
|
||||
points = np.arange(5, dtype=point_type)
|
||||
result = k(points)
|
||||
# weights are always cast to float64
|
||||
assert result.dtype == np.result_type(dataset, points, np.float64(weights),
|
||||
k.factor)
|
||||
|
||||
|
||||
def test_pdf_logpdf():
|
||||
np.random.seed(1)
|
||||
n_basesample = 50
|
||||
xn = np.random.randn(n_basesample)
|
||||
|
||||
# Default
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
|
||||
xs = np.linspace(-15, 12, 25)
|
||||
pdf = gkde.evaluate(xs)
|
||||
pdf2 = gkde.pdf(xs)
|
||||
assert_almost_equal(pdf, pdf2, decimal=12)
|
||||
|
||||
logpdf = np.log(pdf)
|
||||
logpdf2 = gkde.logpdf(xs)
|
||||
assert_almost_equal(logpdf, logpdf2, decimal=12)
|
||||
|
||||
# There are more points than data
|
||||
gkde = stats.gaussian_kde(xs)
|
||||
pdf = np.log(gkde.evaluate(xn))
|
||||
pdf2 = gkde.logpdf(xn)
|
||||
assert_almost_equal(pdf, pdf2, decimal=12)
|
||||
|
||||
|
||||
def test_pdf_logpdf_weighted():
|
||||
np.random.seed(1)
|
||||
n_basesample = 50
|
||||
xn = np.random.randn(n_basesample)
|
||||
wn = np.random.rand(n_basesample)
|
||||
|
||||
# Default
|
||||
gkde = stats.gaussian_kde(xn, weights=wn)
|
||||
|
||||
xs = np.linspace(-15, 12, 25)
|
||||
pdf = gkde.evaluate(xs)
|
||||
pdf2 = gkde.pdf(xs)
|
||||
assert_almost_equal(pdf, pdf2, decimal=12)
|
||||
|
||||
logpdf = np.log(pdf)
|
||||
logpdf2 = gkde.logpdf(xs)
|
||||
assert_almost_equal(logpdf, logpdf2, decimal=12)
|
||||
|
||||
# There are more points than data
|
||||
gkde = stats.gaussian_kde(xs, weights=np.random.rand(len(xs)))
|
||||
pdf = np.log(gkde.evaluate(xn))
|
||||
pdf2 = gkde.logpdf(xn)
|
||||
assert_almost_equal(pdf, pdf2, decimal=12)
|
||||
|
||||
|
||||
@pytest.mark.xslow
|
||||
def test_logpdf_overflow():
|
||||
# regression test for gh-12988; testing against linalg instability for
|
||||
# very high dimensionality kde
|
||||
np.random.seed(1)
|
||||
n_dimensions = 2500
|
||||
n_samples = 5000
|
||||
xn = np.array([np.random.randn(n_samples) + (n) for n in range(
|
||||
0, n_dimensions)])
|
||||
|
||||
# Default
|
||||
gkde = stats.gaussian_kde(xn)
|
||||
|
||||
logpdf = gkde.logpdf(np.arange(0, n_dimensions))
|
||||
np.testing.assert_equal(np.isneginf(logpdf[0]), False)
|
||||
np.testing.assert_equal(np.isnan(logpdf[0]), False)
|
||||
|
||||
|
||||
def test_weights_intact():
|
||||
# regression test for gh-9709: weights are not modified
|
||||
np.random.seed(12345)
|
||||
vals = np.random.lognormal(size=100)
|
||||
weights = np.random.choice([1.0, 10.0, 100], size=vals.size)
|
||||
orig_weights = weights.copy()
|
||||
|
||||
stats.gaussian_kde(np.log10(vals), weights=weights)
|
||||
assert_allclose(weights, orig_weights, atol=1e-14, rtol=1e-14)
|
||||
|
||||
|
||||
def test_weights_integer():
|
||||
# integer weights are OK, cf gh-9709 (comment)
|
||||
np.random.seed(12345)
|
||||
values = [0.2, 13.5, 21.0, 75.0, 99.0]
|
||||
weights = [1, 2, 4, 8, 16] # a list of integers
|
||||
pdf_i = stats.gaussian_kde(values, weights=weights)
|
||||
pdf_f = stats.gaussian_kde(values, weights=np.float64(weights))
|
||||
|
||||
xn = [0.3, 11, 88]
|
||||
assert_allclose(pdf_i.evaluate(xn),
|
||||
pdf_f.evaluate(xn), atol=1e-14, rtol=1e-14)
|
||||
|
||||
|
||||
def test_seed():
|
||||
# Test the seed option of the resample method
|
||||
def test_seed_sub(gkde_trail):
|
||||
n_sample = 200
|
||||
# The results should be different without using seed
|
||||
samp1 = gkde_trail.resample(n_sample)
|
||||
samp2 = gkde_trail.resample(n_sample)
|
||||
assert_raises(
|
||||
AssertionError, assert_allclose, samp1, samp2, atol=1e-13
|
||||
)
|
||||
# Use integer seed
|
||||
seed = 831
|
||||
samp1 = gkde_trail.resample(n_sample, seed=seed)
|
||||
samp2 = gkde_trail.resample(n_sample, seed=seed)
|
||||
assert_allclose(samp1, samp2, atol=1e-13)
|
||||
# Use RandomState
|
||||
rstate1 = np.random.RandomState(seed=138)
|
||||
samp1 = gkde_trail.resample(n_sample, seed=rstate1)
|
||||
rstate2 = np.random.RandomState(seed=138)
|
||||
samp2 = gkde_trail.resample(n_sample, seed=rstate2)
|
||||
assert_allclose(samp1, samp2, atol=1e-13)
|
||||
|
||||
# check that np.random.Generator can be used (numpy >= 1.17)
|
||||
if hasattr(np.random, 'default_rng'):
|
||||
# obtain a np.random.Generator object
|
||||
rng = np.random.default_rng(1234)
|
||||
gkde_trail.resample(n_sample, seed=rng)
|
||||
|
||||
np.random.seed(8765678)
|
||||
n_basesample = 500
|
||||
wn = np.random.rand(n_basesample)
|
||||
# Test 1D case
|
||||
xn_1d = np.random.randn(n_basesample)
|
||||
|
||||
gkde_1d = stats.gaussian_kde(xn_1d)
|
||||
test_seed_sub(gkde_1d)
|
||||
gkde_1d_weighted = stats.gaussian_kde(xn_1d, weights=wn)
|
||||
test_seed_sub(gkde_1d_weighted)
|
||||
|
||||
# Test 2D case
|
||||
mean = np.array([1.0, 3.0])
|
||||
covariance = np.array([[1.0, 2.0], [2.0, 6.0]])
|
||||
xn_2d = np.random.multivariate_normal(mean, covariance, size=n_basesample).T
|
||||
|
||||
gkde_2d = stats.gaussian_kde(xn_2d)
|
||||
test_seed_sub(gkde_2d)
|
||||
gkde_2d_weighted = stats.gaussian_kde(xn_2d, weights=wn)
|
||||
test_seed_sub(gkde_2d_weighted)
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,134 @@
|
||||
import numpy as np
|
||||
import numpy.ma as ma
|
||||
import scipy.stats.mstats as ms
|
||||
|
||||
from numpy.testing import (assert_equal, assert_almost_equal, assert_,
|
||||
assert_allclose)
|
||||
|
||||
|
||||
def test_compare_medians_ms():
|
||||
x = np.arange(7)
|
||||
y = x + 10
|
||||
assert_almost_equal(ms.compare_medians_ms(x, y), 0)
|
||||
|
||||
y2 = np.linspace(0, 1, num=10)
|
||||
assert_almost_equal(ms.compare_medians_ms(x, y2), 0.017116406778)
|
||||
|
||||
|
||||
def test_hdmedian():
|
||||
# 1-D array
|
||||
x = ma.arange(11)
|
||||
assert_allclose(ms.hdmedian(x), 5, rtol=1e-14)
|
||||
x.mask = ma.make_mask(x)
|
||||
x.mask[:7] = False
|
||||
assert_allclose(ms.hdmedian(x), 3, rtol=1e-14)
|
||||
|
||||
# Check that `var` keyword returns a value. TODO: check whether returned
|
||||
# value is actually correct.
|
||||
assert_(ms.hdmedian(x, var=True).size == 2)
|
||||
|
||||
# 2-D array
|
||||
x2 = ma.arange(22).reshape((11, 2))
|
||||
assert_allclose(ms.hdmedian(x2, axis=0), [10, 11])
|
||||
x2.mask = ma.make_mask(x2)
|
||||
x2.mask[:7, :] = False
|
||||
assert_allclose(ms.hdmedian(x2, axis=0), [6, 7])
|
||||
|
||||
|
||||
def test_rsh():
|
||||
np.random.seed(132345)
|
||||
x = np.random.randn(100)
|
||||
res = ms.rsh(x)
|
||||
# Just a sanity check that the code runs and output shape is correct.
|
||||
# TODO: check that implementation is correct.
|
||||
assert_(res.shape == x.shape)
|
||||
|
||||
# Check points keyword
|
||||
res = ms.rsh(x, points=[0, 1.])
|
||||
assert_(res.size == 2)
|
||||
|
||||
|
||||
def test_mjci():
|
||||
# Tests the Marits-Jarrett estimator
|
||||
data = ma.array([77, 87, 88,114,151,210,219,246,253,262,
|
||||
296,299,306,376,428,515,666,1310,2611])
|
||||
assert_almost_equal(ms.mjci(data),[55.76819,45.84028,198.87875],5)
|
||||
|
||||
|
||||
def test_trimmed_mean_ci():
|
||||
# Tests the confidence intervals of the trimmed mean.
|
||||
data = ma.array([545,555,558,572,575,576,578,580,
|
||||
594,605,635,651,653,661,666])
|
||||
assert_almost_equal(ms.trimmed_mean(data,0.2), 596.2, 1)
|
||||
assert_equal(np.round(ms.trimmed_mean_ci(data,(0.2,0.2)),1),
|
||||
[561.8, 630.6])
|
||||
|
||||
|
||||
def test_idealfourths():
|
||||
# Tests ideal-fourths
|
||||
test = np.arange(100)
|
||||
assert_almost_equal(np.asarray(ms.idealfourths(test)),
|
||||
[24.416667,74.583333],6)
|
||||
test_2D = test.repeat(3).reshape(-1,3)
|
||||
assert_almost_equal(ms.idealfourths(test_2D, axis=0),
|
||||
[[24.416667,24.416667,24.416667],
|
||||
[74.583333,74.583333,74.583333]],6)
|
||||
assert_almost_equal(ms.idealfourths(test_2D, axis=1),
|
||||
test.repeat(2).reshape(-1,2))
|
||||
test = [0, 0]
|
||||
_result = ms.idealfourths(test)
|
||||
assert_(np.isnan(_result).all())
|
||||
|
||||
|
||||
class TestQuantiles:
|
||||
data = [0.706560797,0.727229578,0.990399276,0.927065621,0.158953014,
|
||||
0.887764025,0.239407086,0.349638551,0.972791145,0.149789972,
|
||||
0.936947700,0.132359948,0.046041972,0.641675031,0.945530547,
|
||||
0.224218684,0.771450991,0.820257774,0.336458052,0.589113496,
|
||||
0.509736129,0.696838829,0.491323573,0.622767425,0.775189248,
|
||||
0.641461450,0.118455200,0.773029450,0.319280007,0.752229111,
|
||||
0.047841438,0.466295911,0.583850781,0.840581845,0.550086491,
|
||||
0.466470062,0.504765074,0.226855960,0.362641207,0.891620942,
|
||||
0.127898691,0.490094097,0.044882048,0.041441695,0.317976349,
|
||||
0.504135618,0.567353033,0.434617473,0.636243375,0.231803616,
|
||||
0.230154113,0.160011327,0.819464108,0.854706985,0.438809221,
|
||||
0.487427267,0.786907310,0.408367937,0.405534192,0.250444460,
|
||||
0.995309248,0.144389588,0.739947527,0.953543606,0.680051621,
|
||||
0.388382017,0.863530727,0.006514031,0.118007779,0.924024803,
|
||||
0.384236354,0.893687694,0.626534881,0.473051932,0.750134705,
|
||||
0.241843555,0.432947602,0.689538104,0.136934797,0.150206859,
|
||||
0.474335206,0.907775349,0.525869295,0.189184225,0.854284286,
|
||||
0.831089744,0.251637345,0.587038213,0.254475554,0.237781276,
|
||||
0.827928620,0.480283781,0.594514455,0.213641488,0.024194386,
|
||||
0.536668589,0.699497811,0.892804071,0.093835427,0.731107772]
|
||||
|
||||
def test_hdquantiles(self):
|
||||
data = self.data
|
||||
assert_almost_equal(ms.hdquantiles(data,[0., 1.]),
|
||||
[0.006514031, 0.995309248])
|
||||
hdq = ms.hdquantiles(data,[0.25, 0.5, 0.75])
|
||||
assert_almost_equal(hdq, [0.253210762, 0.512847491, 0.762232442,])
|
||||
hdq = ms.hdquantiles_sd(data,[0.25, 0.5, 0.75])
|
||||
assert_almost_equal(hdq, [0.03786954, 0.03805389, 0.03800152,], 4)
|
||||
|
||||
data = np.array(data).reshape(10,10)
|
||||
hdq = ms.hdquantiles(data,[0.25,0.5,0.75],axis=0)
|
||||
assert_almost_equal(hdq[:,0], ms.hdquantiles(data[:,0],[0.25,0.5,0.75]))
|
||||
assert_almost_equal(hdq[:,-1], ms.hdquantiles(data[:,-1],[0.25,0.5,0.75]))
|
||||
hdq = ms.hdquantiles(data,[0.25,0.5,0.75],axis=0,var=True)
|
||||
assert_almost_equal(hdq[...,0],
|
||||
ms.hdquantiles(data[:,0],[0.25,0.5,0.75],var=True))
|
||||
assert_almost_equal(hdq[...,-1],
|
||||
ms.hdquantiles(data[:,-1],[0.25,0.5,0.75], var=True))
|
||||
|
||||
def test_hdquantiles_sd(self):
|
||||
# Only test that code runs, implementation not checked for correctness
|
||||
res = ms.hdquantiles_sd(self.data)
|
||||
assert_(res.size == 3)
|
||||
|
||||
def test_mquantiles_cimj(self):
|
||||
# Only test that code runs, implementation not checked for correctness
|
||||
ci_lower, ci_upper = ms.mquantiles_cimj(self.data)
|
||||
assert_(ci_lower.size == ci_upper.size == 3)
|
||||
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,241 @@
|
||||
import numpy as np
|
||||
from numpy.testing import assert_equal, assert_array_equal
|
||||
|
||||
from scipy.stats import rankdata, tiecorrect
|
||||
import pytest
|
||||
|
||||
|
||||
class TestTieCorrect:
|
||||
|
||||
def test_empty(self):
|
||||
"""An empty array requires no correction, should return 1.0."""
|
||||
ranks = np.array([], dtype=np.float64)
|
||||
c = tiecorrect(ranks)
|
||||
assert_equal(c, 1.0)
|
||||
|
||||
def test_one(self):
|
||||
"""A single element requires no correction, should return 1.0."""
|
||||
ranks = np.array([1.0], dtype=np.float64)
|
||||
c = tiecorrect(ranks)
|
||||
assert_equal(c, 1.0)
|
||||
|
||||
def test_no_correction(self):
|
||||
"""Arrays with no ties require no correction."""
|
||||
ranks = np.arange(2.0)
|
||||
c = tiecorrect(ranks)
|
||||
assert_equal(c, 1.0)
|
||||
ranks = np.arange(3.0)
|
||||
c = tiecorrect(ranks)
|
||||
assert_equal(c, 1.0)
|
||||
|
||||
def test_basic(self):
|
||||
"""Check a few basic examples of the tie correction factor."""
|
||||
# One tie of two elements
|
||||
ranks = np.array([1.0, 2.5, 2.5])
|
||||
c = tiecorrect(ranks)
|
||||
T = 2.0
|
||||
N = ranks.size
|
||||
expected = 1.0 - (T**3 - T) / (N**3 - N)
|
||||
assert_equal(c, expected)
|
||||
|
||||
# One tie of two elements (same as above, but tie is not at the end)
|
||||
ranks = np.array([1.5, 1.5, 3.0])
|
||||
c = tiecorrect(ranks)
|
||||
T = 2.0
|
||||
N = ranks.size
|
||||
expected = 1.0 - (T**3 - T) / (N**3 - N)
|
||||
assert_equal(c, expected)
|
||||
|
||||
# One tie of three elements
|
||||
ranks = np.array([1.0, 3.0, 3.0, 3.0])
|
||||
c = tiecorrect(ranks)
|
||||
T = 3.0
|
||||
N = ranks.size
|
||||
expected = 1.0 - (T**3 - T) / (N**3 - N)
|
||||
assert_equal(c, expected)
|
||||
|
||||
# Two ties, lengths 2 and 3.
|
||||
ranks = np.array([1.5, 1.5, 4.0, 4.0, 4.0])
|
||||
c = tiecorrect(ranks)
|
||||
T1 = 2.0
|
||||
T2 = 3.0
|
||||
N = ranks.size
|
||||
expected = 1.0 - ((T1**3 - T1) + (T2**3 - T2)) / (N**3 - N)
|
||||
assert_equal(c, expected)
|
||||
|
||||
def test_overflow(self):
|
||||
ntie, k = 2000, 5
|
||||
a = np.repeat(np.arange(k), ntie)
|
||||
n = a.size # ntie * k
|
||||
out = tiecorrect(rankdata(a))
|
||||
assert_equal(out, 1.0 - k * (ntie**3 - ntie) / float(n**3 - n))
|
||||
|
||||
|
||||
class TestRankData:
|
||||
|
||||
def test_empty(self):
|
||||
"""stats.rankdata([]) should return an empty array."""
|
||||
a = np.array([], dtype=int)
|
||||
r = rankdata(a)
|
||||
assert_array_equal(r, np.array([], dtype=np.float64))
|
||||
r = rankdata([])
|
||||
assert_array_equal(r, np.array([], dtype=np.float64))
|
||||
|
||||
def test_one(self):
|
||||
"""Check stats.rankdata with an array of length 1."""
|
||||
data = [100]
|
||||
a = np.array(data, dtype=int)
|
||||
r = rankdata(a)
|
||||
assert_array_equal(r, np.array([1.0], dtype=np.float64))
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, np.array([1.0], dtype=np.float64))
|
||||
|
||||
def test_basic(self):
|
||||
"""Basic tests of stats.rankdata."""
|
||||
data = [100, 10, 50]
|
||||
expected = np.array([3.0, 1.0, 2.0], dtype=np.float64)
|
||||
a = np.array(data, dtype=int)
|
||||
r = rankdata(a)
|
||||
assert_array_equal(r, expected)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, expected)
|
||||
|
||||
data = [40, 10, 30, 10, 50]
|
||||
expected = np.array([4.0, 1.5, 3.0, 1.5, 5.0], dtype=np.float64)
|
||||
a = np.array(data, dtype=int)
|
||||
r = rankdata(a)
|
||||
assert_array_equal(r, expected)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, expected)
|
||||
|
||||
data = [20, 20, 20, 10, 10, 10]
|
||||
expected = np.array([5.0, 5.0, 5.0, 2.0, 2.0, 2.0], dtype=np.float64)
|
||||
a = np.array(data, dtype=int)
|
||||
r = rankdata(a)
|
||||
assert_array_equal(r, expected)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, expected)
|
||||
# The docstring states explicitly that the argument is flattened.
|
||||
a2d = a.reshape(2, 3)
|
||||
r = rankdata(a2d)
|
||||
assert_array_equal(r, expected)
|
||||
|
||||
def test_rankdata_object_string(self):
|
||||
min_rank = lambda a: [1 + sum(i < j for i in a) for j in a]
|
||||
max_rank = lambda a: [sum(i <= j for i in a) for j in a]
|
||||
ordinal_rank = lambda a: min_rank([(x, i) for i, x in enumerate(a)])
|
||||
|
||||
def average_rank(a):
|
||||
return [(i + j) / 2.0 for i, j in zip(min_rank(a), max_rank(a))]
|
||||
|
||||
def dense_rank(a):
|
||||
b = np.unique(a)
|
||||
return [1 + sum(i < j for i in b) for j in a]
|
||||
|
||||
rankf = dict(min=min_rank, max=max_rank, ordinal=ordinal_rank,
|
||||
average=average_rank, dense=dense_rank)
|
||||
|
||||
def check_ranks(a):
|
||||
for method in 'min', 'max', 'dense', 'ordinal', 'average':
|
||||
out = rankdata(a, method=method)
|
||||
assert_array_equal(out, rankf[method](a))
|
||||
|
||||
val = ['foo', 'bar', 'qux', 'xyz', 'abc', 'efg', 'ace', 'qwe', 'qaz']
|
||||
check_ranks(np.random.choice(val, 200))
|
||||
check_ranks(np.random.choice(val, 200).astype('object'))
|
||||
|
||||
val = np.array([0, 1, 2, 2.718, 3, 3.141], dtype='object')
|
||||
check_ranks(np.random.choice(val, 200).astype('object'))
|
||||
|
||||
def test_large_int(self):
|
||||
data = np.array([2**60, 2**60+1], dtype=np.uint64)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, [1.0, 2.0])
|
||||
|
||||
data = np.array([2**60, 2**60+1], dtype=np.int64)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, [1.0, 2.0])
|
||||
|
||||
data = np.array([2**60, -2**60+1], dtype=np.int64)
|
||||
r = rankdata(data)
|
||||
assert_array_equal(r, [2.0, 1.0])
|
||||
|
||||
def test_big_tie(self):
|
||||
for n in [10000, 100000, 1000000]:
|
||||
data = np.ones(n, dtype=int)
|
||||
r = rankdata(data)
|
||||
expected_rank = 0.5 * (n + 1)
|
||||
assert_array_equal(r, expected_rank * data,
|
||||
"test failed with n=%d" % n)
|
||||
|
||||
def test_axis(self):
|
||||
data = [[0, 2, 1],
|
||||
[4, 2, 2]]
|
||||
expected0 = [[1., 1.5, 1.],
|
||||
[2., 1.5, 2.]]
|
||||
r0 = rankdata(data, axis=0)
|
||||
assert_array_equal(r0, expected0)
|
||||
expected1 = [[1., 3., 2.],
|
||||
[3., 1.5, 1.5]]
|
||||
r1 = rankdata(data, axis=1)
|
||||
assert_array_equal(r1, expected1)
|
||||
|
||||
methods = ["average", "min", "max", "dense", "ordinal"]
|
||||
dtypes = [np.float64] + [np.int_]*4
|
||||
|
||||
@pytest.mark.parametrize("axis", [0, 1])
|
||||
@pytest.mark.parametrize("method, dtype", zip(methods, dtypes))
|
||||
def test_size_0_axis(self, axis, method, dtype):
|
||||
shape = (3, 0)
|
||||
data = np.zeros(shape)
|
||||
r = rankdata(data, method=method, axis=axis)
|
||||
assert_equal(r.shape, shape)
|
||||
assert_equal(r.dtype, dtype)
|
||||
|
||||
|
||||
_cases = (
|
||||
# values, method, expected
|
||||
([], 'average', []),
|
||||
([], 'min', []),
|
||||
([], 'max', []),
|
||||
([], 'dense', []),
|
||||
([], 'ordinal', []),
|
||||
#
|
||||
([100], 'average', [1.0]),
|
||||
([100], 'min', [1.0]),
|
||||
([100], 'max', [1.0]),
|
||||
([100], 'dense', [1.0]),
|
||||
([100], 'ordinal', [1.0]),
|
||||
#
|
||||
([100, 100, 100], 'average', [2.0, 2.0, 2.0]),
|
||||
([100, 100, 100], 'min', [1.0, 1.0, 1.0]),
|
||||
([100, 100, 100], 'max', [3.0, 3.0, 3.0]),
|
||||
([100, 100, 100], 'dense', [1.0, 1.0, 1.0]),
|
||||
([100, 100, 100], 'ordinal', [1.0, 2.0, 3.0]),
|
||||
#
|
||||
([100, 300, 200], 'average', [1.0, 3.0, 2.0]),
|
||||
([100, 300, 200], 'min', [1.0, 3.0, 2.0]),
|
||||
([100, 300, 200], 'max', [1.0, 3.0, 2.0]),
|
||||
([100, 300, 200], 'dense', [1.0, 3.0, 2.0]),
|
||||
([100, 300, 200], 'ordinal', [1.0, 3.0, 2.0]),
|
||||
#
|
||||
([100, 200, 300, 200], 'average', [1.0, 2.5, 4.0, 2.5]),
|
||||
([100, 200, 300, 200], 'min', [1.0, 2.0, 4.0, 2.0]),
|
||||
([100, 200, 300, 200], 'max', [1.0, 3.0, 4.0, 3.0]),
|
||||
([100, 200, 300, 200], 'dense', [1.0, 2.0, 3.0, 2.0]),
|
||||
([100, 200, 300, 200], 'ordinal', [1.0, 2.0, 4.0, 3.0]),
|
||||
#
|
||||
([100, 200, 300, 200, 100], 'average', [1.5, 3.5, 5.0, 3.5, 1.5]),
|
||||
([100, 200, 300, 200, 100], 'min', [1.0, 3.0, 5.0, 3.0, 1.0]),
|
||||
([100, 200, 300, 200, 100], 'max', [2.0, 4.0, 5.0, 4.0, 2.0]),
|
||||
([100, 200, 300, 200, 100], 'dense', [1.0, 2.0, 3.0, 2.0, 1.0]),
|
||||
([100, 200, 300, 200, 100], 'ordinal', [1.0, 3.0, 5.0, 4.0, 2.0]),
|
||||
#
|
||||
([10] * 30, 'ordinal', np.arange(1.0, 31.0)),
|
||||
)
|
||||
|
||||
|
||||
def test_cases():
|
||||
for values, method, expected in _cases:
|
||||
r = rankdata(values, method=method)
|
||||
assert_array_equal(r, expected)
|
||||
@@ -0,0 +1,96 @@
|
||||
|
||||
import pytest
|
||||
import numpy as np
|
||||
from numpy.testing import assert_allclose, assert_equal
|
||||
from scipy.stats.contingency import relative_risk
|
||||
|
||||
|
||||
# Test just the calculation of the relative risk, including edge
|
||||
# cases that result in a relative risk of 0, inf or nan.
|
||||
@pytest.mark.parametrize(
|
||||
'exposed_cases, exposed_total, control_cases, control_total, expected_rr',
|
||||
[(1, 4, 3, 8, 0.25 / 0.375),
|
||||
(0, 10, 5, 20, 0),
|
||||
(0, 10, 0, 20, np.nan),
|
||||
(5, 15, 0, 20, np.inf)]
|
||||
)
|
||||
def test_relative_risk(exposed_cases, exposed_total,
|
||||
control_cases, control_total, expected_rr):
|
||||
result = relative_risk(exposed_cases, exposed_total,
|
||||
control_cases, control_total)
|
||||
assert_allclose(result.relative_risk, expected_rr, rtol=1e-13)
|
||||
|
||||
|
||||
def test_relative_risk_confidence_interval():
|
||||
result = relative_risk(exposed_cases=16, exposed_total=128,
|
||||
control_cases=24, control_total=256)
|
||||
rr = result.relative_risk
|
||||
ci = result.confidence_interval(confidence_level=0.95)
|
||||
# The corresponding calculation in R using the epitools package.
|
||||
#
|
||||
# > library(epitools)
|
||||
# > c <- matrix(c(232, 112, 24, 16), nrow=2)
|
||||
# > result <- riskratio(c)
|
||||
# > result$measure
|
||||
# risk ratio with 95% C.I.
|
||||
# Predictor estimate lower upper
|
||||
# Exposed1 1.000000 NA NA
|
||||
# Exposed2 1.333333 0.7347317 2.419628
|
||||
#
|
||||
# The last line is the result that we want.
|
||||
assert_allclose(rr, 4/3)
|
||||
assert_allclose((ci.low, ci.high), (0.7347317, 2.419628), rtol=5e-7)
|
||||
|
||||
|
||||
def test_relative_risk_ci_conflevel0():
|
||||
result = relative_risk(exposed_cases=4, exposed_total=12,
|
||||
control_cases=5, control_total=30)
|
||||
rr = result.relative_risk
|
||||
assert_allclose(rr, 2.0, rtol=1e-14)
|
||||
ci = result.confidence_interval(0)
|
||||
assert_allclose((ci.low, ci.high), (2.0, 2.0), rtol=1e-12)
|
||||
|
||||
|
||||
def test_relative_risk_ci_conflevel1():
|
||||
result = relative_risk(exposed_cases=4, exposed_total=12,
|
||||
control_cases=5, control_total=30)
|
||||
ci = result.confidence_interval(1)
|
||||
assert_equal((ci.low, ci.high), (0, np.inf))
|
||||
|
||||
|
||||
def test_relative_risk_ci_edge_cases_00():
|
||||
result = relative_risk(exposed_cases=0, exposed_total=12,
|
||||
control_cases=0, control_total=30)
|
||||
assert_equal(result.relative_risk, np.nan)
|
||||
ci = result.confidence_interval()
|
||||
assert_equal((ci.low, ci.high), (np.nan, np.nan))
|
||||
|
||||
|
||||
def test_relative_risk_ci_edge_cases_01():
|
||||
result = relative_risk(exposed_cases=0, exposed_total=12,
|
||||
control_cases=1, control_total=30)
|
||||
assert_equal(result.relative_risk, 0)
|
||||
ci = result.confidence_interval()
|
||||
assert_equal((ci.low, ci.high), (0.0, np.nan))
|
||||
|
||||
|
||||
def test_relative_risk_ci_edge_cases_10():
|
||||
result = relative_risk(exposed_cases=1, exposed_total=12,
|
||||
control_cases=0, control_total=30)
|
||||
assert_equal(result.relative_risk, np.inf)
|
||||
ci = result.confidence_interval()
|
||||
assert_equal((ci.low, ci.high), (np.nan, np.inf))
|
||||
|
||||
|
||||
@pytest.mark.parametrize('ec, et, cc, ct', [(0, 0, 10, 20),
|
||||
(-1, 10, 1, 5),
|
||||
(1, 10, 0, 0),
|
||||
(1, 10, -1, 4)])
|
||||
def test_relative_risk_bad_value(ec, et, cc, ct):
|
||||
with pytest.raises(ValueError, match="must be an integer not less than"):
|
||||
relative_risk(ec, et, cc, ct)
|
||||
|
||||
|
||||
def test_relative_risk_bad_type():
|
||||
with pytest.raises(TypeError, match="must be an integer"):
|
||||
relative_risk(1, 10, 2.0, 40)
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,86 @@
|
||||
import numpy as np
|
||||
from numpy.testing import assert_allclose, assert_equal
|
||||
|
||||
from scipy.stats._tukeylambda_stats import (tukeylambda_variance,
|
||||
tukeylambda_kurtosis)
|
||||
|
||||
|
||||
def test_tukeylambda_stats_known_exact():
|
||||
"""Compare results with some known exact formulas."""
|
||||
# Some exact values of the Tukey Lambda variance and kurtosis:
|
||||
# lambda var kurtosis
|
||||
# 0 pi**2/3 6/5 (logistic distribution)
|
||||
# 0.5 4 - pi (5/3 - pi/2)/(pi/4 - 1)**2 - 3
|
||||
# 1 1/3 -6/5 (uniform distribution on (-1,1))
|
||||
# 2 1/12 -6/5 (uniform distribution on (-1/2, 1/2))
|
||||
|
||||
# lambda = 0
|
||||
var = tukeylambda_variance(0)
|
||||
assert_allclose(var, np.pi**2 / 3, atol=1e-12)
|
||||
kurt = tukeylambda_kurtosis(0)
|
||||
assert_allclose(kurt, 1.2, atol=1e-10)
|
||||
|
||||
# lambda = 0.5
|
||||
var = tukeylambda_variance(0.5)
|
||||
assert_allclose(var, 4 - np.pi, atol=1e-12)
|
||||
kurt = tukeylambda_kurtosis(0.5)
|
||||
desired = (5./3 - np.pi/2) / (np.pi/4 - 1)**2 - 3
|
||||
assert_allclose(kurt, desired, atol=1e-10)
|
||||
|
||||
# lambda = 1
|
||||
var = tukeylambda_variance(1)
|
||||
assert_allclose(var, 1.0 / 3, atol=1e-12)
|
||||
kurt = tukeylambda_kurtosis(1)
|
||||
assert_allclose(kurt, -1.2, atol=1e-10)
|
||||
|
||||
# lambda = 2
|
||||
var = tukeylambda_variance(2)
|
||||
assert_allclose(var, 1.0 / 12, atol=1e-12)
|
||||
kurt = tukeylambda_kurtosis(2)
|
||||
assert_allclose(kurt, -1.2, atol=1e-10)
|
||||
|
||||
|
||||
def test_tukeylambda_stats_mpmath():
|
||||
"""Compare results with some values that were computed using mpmath."""
|
||||
a10 = dict(atol=1e-10, rtol=0)
|
||||
a12 = dict(atol=1e-12, rtol=0)
|
||||
data = [
|
||||
# lambda variance kurtosis
|
||||
[-0.1, 4.78050217874253547, 3.78559520346454510],
|
||||
[-0.0649, 4.16428023599895777, 2.52019675947435718],
|
||||
[-0.05, 3.93672267890775277, 2.13129793057777277],
|
||||
[-0.001, 3.30128380390964882, 1.21452460083542988],
|
||||
[0.001, 3.27850775649572176, 1.18560634779287585],
|
||||
[0.03125, 2.95927803254615800, 0.804487555161819980],
|
||||
[0.05, 2.78281053405464501, 0.611604043886644327],
|
||||
[0.0649, 2.65282386754100551, 0.476834119532774540],
|
||||
[1.2, 0.242153920578588346, -1.23428047169049726],
|
||||
[10.0, 0.00095237579757703597, 2.37810697355144933],
|
||||
[20.0, 0.00012195121951131043, 7.37654321002709531],
|
||||
]
|
||||
|
||||
for lam, var_expected, kurt_expected in data:
|
||||
var = tukeylambda_variance(lam)
|
||||
assert_allclose(var, var_expected, **a12)
|
||||
kurt = tukeylambda_kurtosis(lam)
|
||||
assert_allclose(kurt, kurt_expected, **a10)
|
||||
|
||||
# Test with vector arguments (most of the other tests are for single
|
||||
# values).
|
||||
lam, var_expected, kurt_expected = zip(*data)
|
||||
var = tukeylambda_variance(lam)
|
||||
assert_allclose(var, var_expected, **a12)
|
||||
kurt = tukeylambda_kurtosis(lam)
|
||||
assert_allclose(kurt, kurt_expected, **a10)
|
||||
|
||||
|
||||
def test_tukeylambda_stats_invalid():
|
||||
"""Test values of lambda outside the domains of the functions."""
|
||||
lam = [-1.0, -0.5]
|
||||
var = tukeylambda_variance(lam)
|
||||
assert_equal(var, np.array([np.nan, np.inf]))
|
||||
|
||||
lam = [-1.0, -0.25]
|
||||
kurt = tukeylambda_kurtosis(lam)
|
||||
assert_equal(kurt, np.array([np.nan, np.inf]))
|
||||
|
||||
@@ -0,0 +1,158 @@
|
||||
import numpy as np
|
||||
from numpy.testing import assert_equal, assert_allclose
|
||||
import pytest
|
||||
from scipy.stats import variation
|
||||
|
||||
|
||||
class TestVariation:
|
||||
"""
|
||||
Test class for scipy.stats.variation
|
||||
"""
|
||||
|
||||
def test_ddof(self):
|
||||
x = np.arange(9.0)
|
||||
assert_allclose(variation(x, ddof=1), np.sqrt(60/8)/4)
|
||||
|
||||
@pytest.mark.parametrize('sgn', [1, -1])
|
||||
def test_sign(self, sgn):
|
||||
x = np.array([1, 2, 3, 4, 5])
|
||||
v = variation(sgn*x)
|
||||
expected = sgn*np.sqrt(2)/3
|
||||
assert_allclose(v, expected, rtol=1e-10)
|
||||
|
||||
def test_scalar(self):
|
||||
# A scalar is treated like a 1-d sequence with length 1.
|
||||
assert_equal(variation(4.0), 0.0)
|
||||
|
||||
@pytest.mark.parametrize('nan_policy, expected',
|
||||
[('propagate', np.nan),
|
||||
('omit', np.sqrt(20/3)/4)])
|
||||
def test_variation_nan(self, nan_policy, expected):
|
||||
x = np.arange(10.)
|
||||
x[9] = np.nan
|
||||
assert_allclose(variation(x, nan_policy=nan_policy), expected)
|
||||
|
||||
def test_nan_policy_raise(self):
|
||||
x = np.array([1.0, 2.0, np.nan, 3.0])
|
||||
with pytest.raises(ValueError, match='input contains nan'):
|
||||
variation(x, nan_policy='raise')
|
||||
|
||||
def test_bad_nan_policy(self):
|
||||
with pytest.raises(ValueError, match='must be one of'):
|
||||
variation([1, 2, 3], nan_policy='foobar')
|
||||
|
||||
def test_keepdims(self):
|
||||
x = np.arange(10).reshape(2, 5)
|
||||
y = variation(x, axis=1, keepdims=True)
|
||||
expected = np.array([[np.sqrt(2)/2],
|
||||
[np.sqrt(2)/7]])
|
||||
assert_allclose(y, expected)
|
||||
|
||||
@pytest.mark.parametrize('axis, expected',
|
||||
[(0, np.empty((1, 0))),
|
||||
(1, np.full((5, 1), fill_value=np.nan))])
|
||||
def test_keepdims_size0(self, axis, expected):
|
||||
x = np.zeros((5, 0))
|
||||
y = variation(x, axis=axis, keepdims=True)
|
||||
assert_equal(y, expected)
|
||||
|
||||
@pytest.mark.parametrize('incr, expected_fill', [(0, np.inf), (1, np.nan)])
|
||||
def test_keepdims_and_ddof_eq_len_plus_incr(self, incr, expected_fill):
|
||||
x = np.array([[1, 1, 2, 2], [1, 2, 3, 3]])
|
||||
y = variation(x, axis=1, ddof=x.shape[1] + incr, keepdims=True)
|
||||
assert_equal(y, np.full((2, 1), fill_value=expected_fill))
|
||||
|
||||
def test_propagate_nan(self):
|
||||
# Check that the shape of the result is the same for inputs
|
||||
# with and without nans, cf gh-5817
|
||||
a = np.arange(8).reshape(2, -1).astype(float)
|
||||
a[1, 0] = np.nan
|
||||
v = variation(a, axis=1, nan_policy="propagate")
|
||||
assert_allclose(v, [np.sqrt(5/4)/1.5, np.nan], atol=1e-15)
|
||||
|
||||
def test_axis_none(self):
|
||||
# Check that `variation` computes the result on the flattened
|
||||
# input when axis is None.
|
||||
y = variation([[0, 1], [2, 3]], axis=None)
|
||||
assert_allclose(y, np.sqrt(5/4)/1.5)
|
||||
|
||||
def test_bad_axis(self):
|
||||
# Check that an invalid axis raises np.AxisError.
|
||||
x = np.array([[1, 2, 3], [4, 5, 6]])
|
||||
with pytest.raises(np.AxisError):
|
||||
variation(x, axis=10)
|
||||
|
||||
def test_mean_zero(self):
|
||||
# Check that `variation` returns inf for a sequence that is not
|
||||
# identically zero but whose mean is zero.
|
||||
x = np.array([10, -3, 1, -4, -4])
|
||||
y = variation(x)
|
||||
assert_equal(y, np.inf)
|
||||
|
||||
x2 = np.array([x, -10*x])
|
||||
y2 = variation(x2, axis=1)
|
||||
assert_equal(y2, [np.inf, np.inf])
|
||||
|
||||
@pytest.mark.parametrize('x', [np.zeros(5), [], [1, 2, np.inf, 9]])
|
||||
def test_return_nan(self, x):
|
||||
# Test some cases where `variation` returns nan.
|
||||
y = variation(x)
|
||||
assert_equal(y, np.nan)
|
||||
|
||||
@pytest.mark.parametrize('axis, expected',
|
||||
[(0, []), (1, [np.nan]*3), (None, np.nan)])
|
||||
def test_2d_size_zero_with_axis(self, axis, expected):
|
||||
x = np.empty((3, 0))
|
||||
y = variation(x, axis=axis)
|
||||
assert_equal(y, expected)
|
||||
|
||||
def test_neg_inf(self):
|
||||
# Edge case that produces -inf: ddof equals the number of non-nan
|
||||
# values, the values are not constant, and the mean is negative.
|
||||
x1 = np.array([-3, -5])
|
||||
assert_equal(variation(x1, ddof=2), -np.inf)
|
||||
|
||||
x2 = np.array([[np.nan, 1, -10, np.nan],
|
||||
[-20, -3, np.nan, np.nan]])
|
||||
assert_equal(variation(x2, axis=1, ddof=2, nan_policy='omit'),
|
||||
[-np.inf, -np.inf])
|
||||
|
||||
@pytest.mark.parametrize("nan_policy", ['propagate', 'omit'])
|
||||
def test_combined_edge_cases(self, nan_policy):
|
||||
x = np.array([[0, 10, np.nan, 1],
|
||||
[0, -5, np.nan, 2],
|
||||
[0, -5, np.nan, 3]])
|
||||
y = variation(x, axis=0, nan_policy=nan_policy)
|
||||
assert_allclose(y, [np.nan, np.inf, np.nan, np.sqrt(2/3)/2])
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
'ddof, expected',
|
||||
[(0, [np.sqrt(1/6), np.sqrt(5/8), np.inf, 0, np.nan, 0.0, np.nan]),
|
||||
(1, [0.5, np.sqrt(5/6), np.inf, 0, np.nan, 0, np.nan]),
|
||||
(2, [np.sqrt(0.5), np.sqrt(5/4), np.inf, np.nan, np.nan, 0, np.nan])]
|
||||
)
|
||||
def test_more_nan_policy_omit_tests(self, ddof, expected):
|
||||
# The slightly strange formatting in the follow array is my attempt to
|
||||
# maintain a clean tabular arrangement of the data while satisfying
|
||||
# the demands of pycodestyle. Currently, E201 and E241 are not
|
||||
# disabled by the `# noqa` annotation.
|
||||
nan = np.nan
|
||||
x = np.array([[1.0, 2.0, nan, 3.0],
|
||||
[0.0, 4.0, 3.0, 1.0],
|
||||
[nan, -.5, 0.5, nan],
|
||||
[nan, 9.0, 9.0, nan],
|
||||
[nan, nan, nan, nan],
|
||||
[3.0, 3.0, 3.0, 3.0],
|
||||
[0.0, 0.0, 0.0, 0.0]])
|
||||
v = variation(x, axis=1, ddof=ddof, nan_policy='omit')
|
||||
assert_allclose(v, expected)
|
||||
|
||||
def test_variation_ddof(self):
|
||||
# test variation with delta degrees of freedom
|
||||
# regression test for gh-13341
|
||||
a = np.array([1, 2, 3, 4, 5])
|
||||
nan_a = np.array([1, 2, 3, np.nan, 4, 5, np.nan])
|
||||
y = variation(a, ddof=1)
|
||||
nan_y = variation(nan_a, nan_policy="omit", ddof=1)
|
||||
assert_allclose(y, np.sqrt(5/2)/3)
|
||||
assert y == nan_y
|
||||
Reference in New Issue
Block a user