import os import numpy as np import numpy.testing as npt from numpy.testing import assert_allclose, assert_equal import pytest from scipy import stats from scipy.optimize import differential_evolution from .test_continuous_basic import distcont from scipy.stats._distn_infrastructure import FitError from scipy.stats._distr_params import distdiscrete from scipy.stats import goodness_of_fit # 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 = [ 'gausshyper', 'genexpon', 'gengamma', 'irwinhall', 'kappa4', 'ksone', 'kstwo', 'ncf', 'ncx2', 'truncexpon', 'tukeylambda', 'vonmises', 'levy_stable', 'trapezoid', 'truncweibull_min', 'studentized_range', ] # these pass but are XSLOW (>1s) mle_Xslow_fits = ['betaprime', 'crystalball', 'exponweib', 'f', 'geninvgauss', 'jf_skew_t', 'recipinvgauss', 'rel_breitwigner', 'vonmises_line'] # The MLE fit method of these distributions doesn't perform well when all # parameters are fit, so test them with the location fixed at 0. mle_use_floc0 = [ 'burr', 'chi', 'chi2', 'mielke', 'pearson3', 'genhalflogistic', 'rdist', 'pareto', 'powerlaw', # distfn.nnlf(est2, rvs) > distfn.nnlf(est1, rvs) otherwise 'powerlognorm', 'wrapcauchy', 'rel_breitwigner', ] 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', 'irwinhall', 'jf_skew_t', 'johnsonsu', 'kappa3', 'ksone', 'kstwo', 'levy', 'levy_l', 'levy_stable', 'loglaplace', 'lomax', 'mielke', 'nakagami', 'ncf', 'nct', 'ncx2', 'pareto', 'powerlognorm', 'powernorm', 'rel_breitwigner', 'skewcauchy', 't', 'trapezoid', 'triang', 'truncpareto', 'truncweibull_min', 'tukeylambda', 'studentized_range'] # not sure if these fail, but they caused my patience to fail mm_XXslow_fits = ['argus', 'exponpow', 'exponweib', 'gausshyper', 'genexpon', 'genhalflogistic', 'halfgennorm', 'gompertz', 'johnsonsb', 'kappa4', 'kstwobign', 'recipinvgauss', 'truncexpon', 'vonmises', 'vonmises_line'] # these pass but are XSLOW (>1s) mm_Xslow_fits = ['wrapcauchy'] failing_fits = {"MM": mm_failing_fits + mm_XXslow_fits, "MLE": mle_failing_fits} xslow_fits = {"MM": mm_Xslow_fits, "MLE": mle_Xslow_fits} fail_interval_censored = {"truncpareto"} # Don't run the fit test on these: skip_fit = [ 'erlang', # Subclass of gamma, generates a warning. 'genhyperbolic', 'norminvgauss', # too slow ] 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): run_xfail = int(os.getenv('SCIPY_XFAIL', default=False)) run_xslow = int(os.getenv('SCIPY_XSLOW', default=False)) if distname in failing_fits[method] and not run_xfail: # The generic `fit` method can't be expected to work perfectly for all # distributions, data, and guesses. Some failures are expected. msg = "Failure expected; set environment variable SCIPY_XFAIL=1 to run." pytest.xfail(msg) if distname in xslow_fits[method] and not run_xslow: msg = "Very slow; set environment variable SCIPY_XSLOW=1 to run." pytest.skip(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) if method == 'MLE' and distfn.name in mle_use_floc0: kwds = {'floc': 0} else: kwds = {} # start with default values est = distfn.fit(rvs, method=method, **kwds) if method == 'MLE': # Trivial test of the use of CensoredData. The fit() method # will check that data contains no actual censored data, and # do a regular uncensored fit. data1 = stats.CensoredData(rvs) est1 = distfn.fit(data1, **kwds) msg = ('Different results fitting uncensored data wrapped as' f' CensoredData: {distfn.name}: est={est} est1={est1}') assert_allclose(est1, est, rtol=1e-10, err_msg=msg) if method == 'MLE' and distname not in fail_interval_censored: # Convert the first `nic` values in rvs to interval-censored # values. The interval is small, so est2 should be close to # est. nic = 15 interval = np.column_stack((rvs, rvs)) interval[:nic, 0] *= 0.99 interval[:nic, 1] *= 1.01 interval.sort(axis=1) data2 = stats.CensoredData(interval=interval) est2 = distfn.fit(data2, **kwds) msg = ('Different results fitting interval-censored' f' data: {distfn.name}: est={est} est2={est2}') assert_allclose(est2, est, rtol=0.05, err_msg=msg) 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) def test_fit_error(): data = np.concatenate([np.zeros(29), np.ones(21)]) message = "Optimization converged to parameters that are..." with pytest.raises(FitError, match=message), \ pytest.warns(RuntimeWarning): stats.beta.fit(data) @pytest.mark.parametrize("dist, params", [(stats.norm, (0.5, 2.5)), # type: ignore[attr-defined] (stats.binom, (10, 0.3, 2))]) # type: ignore[attr-defined] def test_nnlf_and_related_methods(dist, params): rng = np.random.default_rng(983459824) if hasattr(dist, 'pdf'): logpxf = dist.logpdf else: logpxf = dist.logpmf x = dist.rvs(*params, size=100, random_state=rng) ref = -logpxf(x, *params).sum() res1 = dist.nnlf(params, x) res2 = dist._penalized_nnlf(params, x) assert_allclose(res1, ref) assert_allclose(res2, ref) def cases_test_fit_mle(): # These fail default test or hang skip_basic_fit = {'argus', 'irwinhall', 'foldnorm', 'truncpareto', 'truncweibull_min', 'ksone', 'levy_stable', 'studentized_range', 'kstwo', 'arcsine'} # Please keep this list in alphabetical order... slow_basic_fit = {'alpha', 'betaprime', 'binom', 'bradford', 'burr12', 'chi', 'crystalball', 'dweibull', 'erlang', 'exponnorm', 'exponpow', 'f', 'fatiguelife', 'fisk', 'foldcauchy', 'gamma', 'genexpon', 'genextreme', 'gennorm', 'genpareto', 'gompertz', 'halfgennorm', 'invgamma', 'invgauss', 'invweibull', 'jf_skew_t', 'johnsonsb', 'johnsonsu', 'kappa3', 'kstwobign', 'loglaplace', 'lognorm', 'lomax', 'mielke', 'nakagami', 'nbinom', 'norminvgauss', 'pareto', 'pearson3', 'powerlaw', 'powernorm', 'randint', 'rdist', 'recipinvgauss', 'rice', 'skewnorm', 't', 'uniform', 'weibull_max', 'weibull_min', 'wrapcauchy'} # Please keep this list in alphabetical order... xslow_basic_fit = {'beta', 'betabinom', 'betanbinom', 'burr', 'exponweib', 'gausshyper', 'gengamma', 'genhalflogistic', 'genhyperbolic', 'geninvgauss', 'hypergeom', 'kappa4', 'loguniform', 'ncf', 'nchypergeom_fisher', 'nchypergeom_wallenius', 'nct', 'ncx2', 'nhypergeom', 'powerlognorm', 'reciprocal', 'rel_breitwigner', 'skellam', 'trapezoid', 'triang', 'truncnorm', 'tukeylambda', 'vonmises', 'zipfian'} for dist in dict(distdiscrete + distcont): if dist in skip_basic_fit or not isinstance(dist, str): reason = "tested separately" yield pytest.param(dist, marks=pytest.mark.skip(reason=reason)) elif dist in slow_basic_fit: reason = "too slow (>= 0.25s)" yield pytest.param(dist, marks=pytest.mark.slow(reason=reason)) elif dist in xslow_basic_fit: reason = "too slow (>= 1.0s)" yield pytest.param(dist, marks=pytest.mark.xslow(reason=reason)) else: yield dist def cases_test_fit_mse(): # the first four are so slow that I'm not sure whether they would pass skip_basic_fit = {'levy_stable', 'studentized_range', 'ksone', 'skewnorm', 'irwinhall', # hangs 'norminvgauss', # super slow (~1 hr) but passes 'kstwo', # very slow (~25 min) but passes 'geninvgauss', # quite slow (~4 minutes) but passes 'gausshyper', 'genhyperbolic', # integration warnings 'tukeylambda', # close, but doesn't meet tolerance 'vonmises', # can have negative CDF; doesn't play nice 'argus'} # doesn't meet tolerance; tested separately # Please keep this list in alphabetical order... slow_basic_fit = {'alpha', 'anglit', 'arcsine', 'betabinom', 'bradford', 'chi', 'chi2', 'crystalball', 'dweibull', 'erlang', 'exponnorm', 'exponpow', 'exponweib', 'fatiguelife', 'fisk', 'foldcauchy', 'foldnorm', 'gamma', 'genexpon', 'genextreme', 'genhalflogistic', 'genlogistic', 'genpareto', 'gompertz', 'hypergeom', 'invweibull', 'johnsonsu', 'kappa3', 'kstwobign', 'laplace_asymmetric', 'loggamma', 'loglaplace', 'lognorm', 'lomax', 'maxwell', 'nhypergeom', 'pareto', 'powernorm', 'randint', 'recipinvgauss', 'semicircular', 't', 'triang', 'truncexpon', 'truncpareto', 'uniform', 'wald', 'weibull_max', 'weibull_min', 'wrapcauchy'} # Please keep this list in alphabetical order... xslow_basic_fit = {'argus', 'beta', 'betaprime', 'burr', 'burr12', 'dgamma', 'f', 'gengamma', 'gennorm', 'halfgennorm', 'invgamma', 'invgauss', 'jf_skew_t', 'johnsonsb', 'kappa4', 'loguniform', 'mielke', 'nakagami', 'ncf', 'nchypergeom_fisher', 'nchypergeom_wallenius', 'nct', 'ncx2', 'pearson3', 'powerlaw', 'powerlognorm', 'rdist', 'reciprocal', 'rel_breitwigner', 'rice', 'trapezoid', 'truncnorm', 'truncweibull_min', 'vonmises_line', 'zipfian'} warns_basic_fit = {'skellam'} # can remove mark after gh-14901 is resolved for dist in dict(distdiscrete + distcont): if dist in skip_basic_fit or not isinstance(dist, str): reason = "Fails. Oh well." yield pytest.param(dist, marks=pytest.mark.skip(reason=reason)) elif dist in slow_basic_fit: reason = "too slow (>= 0.25s)" yield pytest.param(dist, marks=pytest.mark.slow(reason=reason)) elif dist in xslow_basic_fit: reason = "too slow (>= 1.0s)" yield pytest.param(dist, marks=pytest.mark.xslow(reason=reason)) elif dist in warns_basic_fit: mark = pytest.mark.filterwarnings('ignore::RuntimeWarning') yield pytest.param(dist, marks=mark) else: yield dist def cases_test_fitstart(): for distname, shapes in dict(distcont).items(): if (not isinstance(distname, str) or distname in {'studentized_range', 'recipinvgauss'}): # slow continue yield distname, shapes @pytest.mark.parametrize('distname, shapes', cases_test_fitstart()) def test_fitstart(distname, shapes): dist = getattr(stats, distname) rng = np.random.default_rng(216342614) data = rng.random(10) with np.errstate(invalid='ignore', divide='ignore'): # irrelevant to test guess = dist._fitstart(data) assert dist._argcheck(*guess[:-2]) def assert_nlff_less_or_close(dist, data, params1, params0, rtol=1e-7, atol=0, nlff_name='nnlf'): nlff = getattr(dist, nlff_name) nlff1 = nlff(params1, data) nlff0 = nlff(params0, data) if not (nlff1 < nlff0): np.testing.assert_allclose(nlff1, nlff0, rtol=rtol, atol=atol) class TestFit: dist = stats.binom # type: ignore[attr-defined] seed = 654634816187 rng = np.random.default_rng(seed) data = stats.binom.rvs(5, 0.5, size=100, random_state=rng) # type: ignore[attr-defined] # noqa: E501 shape_bounds_a = [(1, 10), (0, 1)] shape_bounds_d = {'n': (1, 10), 'p': (0, 1)} atol = 5e-2 rtol = 1e-2 tols = {'atol': atol, 'rtol': rtol} def opt(self, *args, **kwds): return differential_evolution(*args, seed=0, **kwds) def test_dist_iv(self): message = "`dist` must be an instance of..." with pytest.raises(ValueError, match=message): stats.fit(10, self.data, self.shape_bounds_a) def test_data_iv(self): message = "`data` must be exactly one-dimensional." with pytest.raises(ValueError, match=message): stats.fit(self.dist, [[1, 2, 3]], self.shape_bounds_a) message = "All elements of `data` must be finite numbers." with pytest.raises(ValueError, match=message): stats.fit(self.dist, [1, 2, 3, np.nan], self.shape_bounds_a) with pytest.raises(ValueError, match=message): stats.fit(self.dist, [1, 2, 3, np.inf], self.shape_bounds_a) with pytest.raises(ValueError, match=message): stats.fit(self.dist, ['1', '2', '3'], self.shape_bounds_a) def test_bounds_iv(self): message = "Bounds provided for the following unrecognized..." shape_bounds = {'n': (1, 10), 'p': (0, 1), '1': (0, 10)} with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "Each element of a `bounds` sequence must be a tuple..." shape_bounds = [(1, 10, 3), (0, 1)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "Each element of `bounds` must be a tuple specifying..." shape_bounds = [(1, 10, 3), (0, 1, 0.5)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) shape_bounds = [1, 0] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "A `bounds` sequence must contain at least 2 elements..." shape_bounds = [(1, 10)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "A `bounds` sequence may not contain more than 3 elements..." bounds = [(1, 10), (1, 10), (1, 10), (1, 10)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, bounds) message = "There are no values for `p` on the interval..." shape_bounds = {'n': (1, 10), 'p': (1, 0)} with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "There are no values for `n` on the interval..." shape_bounds = [(10, 1), (0, 1)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "There are no integer values for `n` on the interval..." shape_bounds = [(1.4, 1.6), (0, 1)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) message = "The intersection of user-provided bounds for `n`" with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data) shape_bounds = [(-np.inf, np.inf), (0, 1)] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, shape_bounds) def test_guess_iv(self): message = "Guesses provided for the following unrecognized..." guess = {'n': 1, 'p': 0.5, '1': 255} with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "Each element of `guess` must be a scalar..." guess = {'n': 1, 'p': 'hi'} with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) guess = [1, 'f'] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) guess = [[1, 2]] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "A `guess` sequence must contain at least 2..." guess = [1] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "A `guess` sequence may not contain more than 3..." guess = [1, 2, 3, 4] with pytest.raises(ValueError, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "Guess for parameter `n` rounded.*|Guess for parameter `p` clipped.*" guess = {'n': 4.5, 'p': -0.5} with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "Guess for parameter `loc` rounded..." guess = [5, 0.5, 0.5] with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "Guess for parameter `p` clipped..." guess = {'n': 5, 'p': -0.5} with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) message = "Guess for parameter `loc` clipped..." guess = [5, 0.5, 1] with pytest.warns(RuntimeWarning, match=message): stats.fit(self.dist, self.data, self.shape_bounds_d, guess=guess) def basic_fit_test(self, dist_name, method): N = 5000 dist_data = dict(distcont + distdiscrete) rng = np.random.default_rng(self.seed) dist = getattr(stats, dist_name) shapes = np.array(dist_data[dist_name]) bounds = np.empty((len(shapes) + 2, 2), dtype=np.float64) bounds[:-2, 0] = shapes/10.**np.sign(shapes) bounds[:-2, 1] = shapes*10.**np.sign(shapes) bounds[-2] = (0, 10) bounds[-1] = (1e-16, 10) loc = rng.uniform(*bounds[-2]) scale = rng.uniform(*bounds[-1]) ref = list(dist_data[dist_name]) + [loc, scale] if getattr(dist, 'pmf', False): ref = ref[:-1] ref[-1] = np.floor(loc) data = dist.rvs(*ref, size=N, random_state=rng) bounds = bounds[:-1] if getattr(dist, 'pdf', False): data = dist.rvs(*ref, size=N, random_state=rng) with npt.suppress_warnings() as sup: sup.filter(RuntimeWarning, "overflow encountered") res = stats.fit(dist, data, bounds, method=method, optimizer=self.opt) nlff_names = {'mle': 'nnlf', 'mse': '_penalized_nlpsf'} nlff_name = nlff_names[method] assert_nlff_less_or_close(dist, data, res.params, ref, **self.tols, nlff_name=nlff_name) @pytest.mark.parametrize("dist_name", cases_test_fit_mle()) def test_basic_fit_mle(self, dist_name): self.basic_fit_test(dist_name, "mle") @pytest.mark.parametrize("dist_name", cases_test_fit_mse()) def test_basic_fit_mse(self, dist_name): self.basic_fit_test(dist_name, "mse") def test_arcsine(self): # Can't guarantee that all distributions will fit all data with # arbitrary bounds. This distribution just happens to fail above. # Try something slightly different. N = 1000 rng = np.random.default_rng(self.seed) dist = stats.arcsine shapes = (1., 2.) data = dist.rvs(*shapes, size=N, random_state=rng) shape_bounds = {'loc': (0.1, 10), 'scale': (0.1, 10)} res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_nlff_less_or_close(dist, data, res.params, shapes, **self.tols) @pytest.mark.parametrize("method", ('mle', 'mse')) def test_argus(self, method): # Can't guarantee that all distributions will fit all data with # arbitrary bounds. This distribution just happens to fail above. # Try something slightly different. N = 1000 rng = np.random.default_rng(self.seed) dist = stats.argus shapes = (1., 2., 3.) data = dist.rvs(*shapes, size=N, random_state=rng) shape_bounds = {'chi': (0.1, 10), 'loc': (0.1, 10), 'scale': (0.1, 10)} res = stats.fit(dist, data, shape_bounds, optimizer=self.opt, method=method) assert_nlff_less_or_close(dist, data, res.params, shapes, **self.tols) def test_foldnorm(self): # Can't guarantee that all distributions will fit all data with # arbitrary bounds. This distribution just happens to fail above. # Try something slightly different. N = 1000 rng = np.random.default_rng(self.seed) dist = stats.foldnorm shapes = (1.952125337355587, 2., 3.) data = dist.rvs(*shapes, size=N, random_state=rng) shape_bounds = {'c': (0.1, 10), 'loc': (0.1, 10), 'scale': (0.1, 10)} res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_nlff_less_or_close(dist, data, res.params, shapes, **self.tols) def test_truncpareto(self): # Can't guarantee that all distributions will fit all data with # arbitrary bounds. This distribution just happens to fail above. # Try something slightly different. N = 1000 rng = np.random.default_rng(self.seed) dist = stats.truncpareto shapes = (1.8, 5.3, 2.3, 4.1) data = dist.rvs(*shapes, size=N, random_state=rng) shape_bounds = [(0.1, 10)]*4 res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_nlff_less_or_close(dist, data, res.params, shapes, **self.tols) def test_truncweibull_min(self): # Can't guarantee that all distributions will fit all data with # arbitrary bounds. This distribution just happens to fail above. # Try something slightly different. N = 1000 rng = np.random.default_rng(self.seed) dist = stats.truncweibull_min shapes = (2.5, 0.25, 1.75, 2., 3.) data = dist.rvs(*shapes, size=N, random_state=rng) shape_bounds = [(0.1, 10)]*5 res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_nlff_less_or_close(dist, data, res.params, shapes, **self.tols) def test_missing_shape_bounds(self): # some distributions have a small domain w.r.t. a parameter, e.g. # $p \in [0, 1]$ for binomial distribution # User does not need to provide these because the intersection of the # user's bounds (none) and the distribution's domain is finite N = 1000 rng = np.random.default_rng(self.seed) dist = stats.binom n, p, loc = 10, 0.65, 0 data = dist.rvs(n, p, loc=loc, size=N, random_state=rng) shape_bounds = {'n': np.array([0, 20])} # check arrays are OK, too res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_allclose(res.params, (n, p, loc), **self.tols) dist = stats.bernoulli p, loc = 0.314159, 0 data = dist.rvs(p, loc=loc, size=N, random_state=rng) res = stats.fit(dist, data, optimizer=self.opt) assert_allclose(res.params, (p, loc), **self.tols) def test_fit_only_loc_scale(self): # fit only loc N = 5000 rng = np.random.default_rng(self.seed) dist = stats.norm loc, scale = 1.5, 1 data = dist.rvs(loc=loc, size=N, random_state=rng) loc_bounds = (0, 5) bounds = {'loc': loc_bounds} res = stats.fit(dist, data, bounds, optimizer=self.opt) assert_allclose(res.params, (loc, scale), **self.tols) # fit only scale loc, scale = 0, 2.5 data = dist.rvs(scale=scale, size=N, random_state=rng) scale_bounds = (0.01, 5) bounds = {'scale': scale_bounds} res = stats.fit(dist, data, bounds, optimizer=self.opt) assert_allclose(res.params, (loc, scale), **self.tols) # fit only loc and scale dist = stats.norm loc, scale = 1.5, 2.5 data = dist.rvs(loc=loc, scale=scale, size=N, random_state=rng) bounds = {'loc': loc_bounds, 'scale': scale_bounds} res = stats.fit(dist, data, bounds, optimizer=self.opt) assert_allclose(res.params, (loc, scale), **self.tols) def test_everything_fixed(self): N = 5000 rng = np.random.default_rng(self.seed) dist = stats.norm loc, scale = 1.5, 2.5 data = dist.rvs(loc=loc, scale=scale, size=N, random_state=rng) # loc, scale fixed to 0, 1 by default res = stats.fit(dist, data) assert_allclose(res.params, (0, 1), **self.tols) # loc, scale explicitly fixed bounds = {'loc': (loc, loc), 'scale': (scale, scale)} res = stats.fit(dist, data, bounds) assert_allclose(res.params, (loc, scale), **self.tols) # `n` gets fixed during polishing dist = stats.binom n, p, loc = 10, 0.65, 0 data = dist.rvs(n, p, loc=loc, size=N, random_state=rng) shape_bounds = {'n': (0, 20), 'p': (0.65, 0.65)} res = stats.fit(dist, data, shape_bounds, optimizer=self.opt) assert_allclose(res.params, (n, p, loc), **self.tols) def test_failure(self): N = 5000 rng = np.random.default_rng(self.seed) dist = stats.nbinom shapes = (5, 0.5) data = dist.rvs(*shapes, size=N, random_state=rng) assert data.min() == 0 # With lower bounds on location at 0.5, likelihood is zero bounds = [(0, 30), (0, 1), (0.5, 10)] res = stats.fit(dist, data, bounds) message = "Optimization converged to parameter values that are" assert res.message.startswith(message) assert res.success is False @pytest.mark.xslow def test_guess(self): # Test that guess helps DE find the desired solution N = 2000 # With some seeds, `fit` doesn't need a guess rng = np.random.default_rng(196390444561) dist = stats.nhypergeom params = (20, 7, 12, 0) bounds = [(2, 200), (0.7, 70), (1.2, 120), (0, 10)] data = dist.rvs(*params, size=N, random_state=rng) res = stats.fit(dist, data, bounds, optimizer=self.opt) assert not np.allclose(res.params, params, **self.tols) res = stats.fit(dist, data, bounds, guess=params, optimizer=self.opt) assert_allclose(res.params, params, **self.tols) def test_mse_accuracy_1(self): # Test maximum spacing estimation against example from Wikipedia # https://en.wikipedia.org/wiki/Maximum_spacing_estimation#Examples data = [2, 4] dist = stats.expon bounds = {'loc': (0, 0), 'scale': (1e-8, 10)} res_mle = stats.fit(dist, data, bounds=bounds, method='mle') assert_allclose(res_mle.params.scale, 3, atol=1e-3) res_mse = stats.fit(dist, data, bounds=bounds, method='mse') assert_allclose(res_mse.params.scale, 3.915, atol=1e-3) def test_mse_accuracy_2(self): # Test maximum spacing estimation against example from Wikipedia # https://en.wikipedia.org/wiki/Maximum_spacing_estimation#Examples rng = np.random.default_rng(9843212616816518964) dist = stats.uniform n = 10 data = dist(3, 6).rvs(size=n, random_state=rng) bounds = {'loc': (0, 10), 'scale': (1e-8, 10)} res = stats.fit(dist, data, bounds=bounds, method='mse') # (loc=3.608118420015416, scale=5.509323262055043) x = np.sort(data) a = (n*x[0] - x[-1])/(n - 1) b = (n*x[-1] - x[0])/(n - 1) ref = a, b-a # (3.6081133632151503, 5.509328130317254) assert_allclose(res.params, ref, rtol=1e-4) # Data from Matlab: https://www.mathworks.com/help/stats/lillietest.html examgrades = [65, 61, 81, 88, 69, 89, 55, 84, 86, 84, 71, 81, 84, 81, 78, 67, 96, 66, 73, 75, 59, 71, 69, 63, 79, 76, 63, 85, 87, 88, 80, 71, 65, 84, 71, 75, 81, 79, 64, 65, 84, 77, 70, 75, 84, 75, 73, 92, 90, 79, 80, 71, 73, 71, 58, 79, 73, 64, 77, 82, 81, 59, 54, 82, 57, 79, 79, 73, 74, 82, 63, 64, 73, 69, 87, 68, 81, 73, 83, 73, 80, 73, 73, 71, 66, 78, 64, 74, 68, 67, 75, 75, 80, 85, 74, 76, 80, 77, 93, 70, 86, 80, 81, 83, 68, 60, 85, 64, 74, 82, 81, 77, 66, 85, 75, 81, 69, 60, 83, 72] class TestGoodnessOfFit: def test_gof_iv(self): dist = stats.norm x = [1, 2, 3] message = r"`dist` must be a \(non-frozen\) instance of..." with pytest.raises(TypeError, match=message): goodness_of_fit(stats.norm(), x) message = "`data` must be a one-dimensional array of numbers." with pytest.raises(ValueError, match=message): goodness_of_fit(dist, [[1, 2, 3]]) message = "`statistic` must be one of..." with pytest.raises(ValueError, match=message): goodness_of_fit(dist, x, statistic='mm') message = "`n_mc_samples` must be an integer." with pytest.raises(TypeError, match=message): goodness_of_fit(dist, x, n_mc_samples=1000.5) message = "'herring' cannot be used to seed a" with pytest.raises(ValueError, match=message): goodness_of_fit(dist, x, random_state='herring') def test_against_ks(self): rng = np.random.default_rng(8517426291317196949) x = examgrades known_params = {'loc': np.mean(x), 'scale': np.std(x, ddof=1)} res = goodness_of_fit(stats.norm, x, known_params=known_params, statistic='ks', random_state=rng) ref = stats.kstest(x, stats.norm(**known_params).cdf, method='exact') assert_allclose(res.statistic, ref.statistic) # ~0.0848 assert_allclose(res.pvalue, ref.pvalue, atol=5e-3) # ~0.335 def test_against_lilliefors(self): rng = np.random.default_rng(2291803665717442724) x = examgrades res = goodness_of_fit(stats.norm, x, statistic='ks', random_state=rng) known_params = {'loc': np.mean(x), 'scale': np.std(x, ddof=1)} ref = stats.kstest(x, stats.norm(**known_params).cdf, method='exact') assert_allclose(res.statistic, ref.statistic) # ~0.0848 assert_allclose(res.pvalue, 0.0348, atol=5e-3) def test_against_cvm(self): rng = np.random.default_rng(8674330857509546614) x = examgrades known_params = {'loc': np.mean(x), 'scale': np.std(x, ddof=1)} res = goodness_of_fit(stats.norm, x, known_params=known_params, statistic='cvm', random_state=rng) ref = stats.cramervonmises(x, stats.norm(**known_params).cdf) assert_allclose(res.statistic, ref.statistic) # ~0.090 assert_allclose(res.pvalue, ref.pvalue, atol=5e-3) # ~0.636 def test_against_anderson_case_0(self): # "Case 0" is where loc and scale are known [1] rng = np.random.default_rng(7384539336846690410) x = np.arange(1, 101) # loc that produced critical value of statistic found w/ root_scalar known_params = {'loc': 45.01575354024957, 'scale': 30} res = goodness_of_fit(stats.norm, x, known_params=known_params, statistic='ad', random_state=rng) assert_allclose(res.statistic, 2.492) # See [1] Table 1A 1.0 assert_allclose(res.pvalue, 0.05, atol=5e-3) def test_against_anderson_case_1(self): # "Case 1" is where scale is known and loc is fit [1] rng = np.random.default_rng(5040212485680146248) x = np.arange(1, 101) # scale that produced critical value of statistic found w/ root_scalar known_params = {'scale': 29.957112639101933} res = goodness_of_fit(stats.norm, x, known_params=known_params, statistic='ad', random_state=rng) assert_allclose(res.statistic, 0.908) # See [1] Table 1B 1.1 assert_allclose(res.pvalue, 0.1, atol=5e-3) def test_against_anderson_case_2(self): # "Case 2" is where loc is known and scale is fit [1] rng = np.random.default_rng(726693985720914083) x = np.arange(1, 101) # loc that produced critical value of statistic found w/ root_scalar known_params = {'loc': 44.5680212261933} res = goodness_of_fit(stats.norm, x, known_params=known_params, statistic='ad', random_state=rng) assert_allclose(res.statistic, 2.904) # See [1] Table 1B 1.2 assert_allclose(res.pvalue, 0.025, atol=5e-3) def test_against_anderson_case_3(self): # "Case 3" is where both loc and scale are fit [1] rng = np.random.default_rng(6763691329830218206) # c that produced critical value of statistic found w/ root_scalar x = stats.skewnorm.rvs(1.4477847789132101, loc=1, scale=2, size=100, random_state=rng) res = goodness_of_fit(stats.norm, x, statistic='ad', random_state=rng) assert_allclose(res.statistic, 0.559) # See [1] Table 1B 1.2 assert_allclose(res.pvalue, 0.15, atol=5e-3) @pytest.mark.xslow def test_against_anderson_gumbel_r(self): rng = np.random.default_rng(7302761058217743) # c that produced critical value of statistic found w/ root_scalar x = stats.genextreme(0.051896837188595134, loc=0.5, scale=1.5).rvs(size=1000, random_state=rng) res = goodness_of_fit(stats.gumbel_r, x, statistic='ad', random_state=rng) ref = stats.anderson(x, dist='gumbel_r') assert_allclose(res.statistic, ref.critical_values[0]) assert_allclose(res.pvalue, ref.significance_level[0]/100, atol=5e-3) def test_against_filliben_norm(self): # Test against `stats.fit` ref. [7] Section 8 "Example" rng = np.random.default_rng(8024266430745011915) y = [6, 1, -4, 8, -2, 5, 0] known_params = {'loc': 0, 'scale': 1} res = stats.goodness_of_fit(stats.norm, y, known_params=known_params, statistic="filliben", random_state=rng) # Slight discrepancy presumably due to roundoff in Filliben's # calculation. Using exact order statistic medians instead of # Filliben's approximation doesn't account for it. assert_allclose(res.statistic, 0.98538, atol=1e-4) assert 0.75 < res.pvalue < 0.9 # Using R's ppcc library: # library(ppcc) # options(digits=16) # x < - c(6, 1, -4, 8, -2, 5, 0) # set.seed(100) # ppccTest(x, "qnorm", ppos="Filliben") # Discrepancy with assert_allclose(res.statistic, 0.98540957187084, rtol=2e-5) assert_allclose(res.pvalue, 0.8875, rtol=2e-3) def test_filliben_property(self): # Filliben's statistic should be independent of data location and scale rng = np.random.default_rng(8535677809395478813) x = rng.normal(loc=10, scale=0.5, size=100) res = stats.goodness_of_fit(stats.norm, x, statistic="filliben", random_state=rng) known_params = {'loc': 0, 'scale': 1} ref = stats.goodness_of_fit(stats.norm, x, known_params=known_params, statistic="filliben", random_state=rng) assert_allclose(res.statistic, ref.statistic, rtol=1e-15) @pytest.mark.parametrize('case', [(25, [.928, .937, .950, .958, .966]), (50, [.959, .965, .972, .977, .981]), (95, [.977, .979, .983, .986, .989])]) def test_against_filliben_norm_table(self, case): # Test against `stats.fit` ref. [7] Table 1 rng = np.random.default_rng(504569995557928957) n, ref = case x = rng.random(n) known_params = {'loc': 0, 'scale': 1} res = stats.goodness_of_fit(stats.norm, x, known_params=known_params, statistic="filliben", random_state=rng) percentiles = np.array([0.005, 0.01, 0.025, 0.05, 0.1]) res = stats.scoreatpercentile(res.null_distribution, percentiles*100) assert_allclose(res, ref, atol=2e-3) @pytest.mark.xslow @pytest.mark.parametrize('case', [(5, 0.95772790260469, 0.4755), (6, 0.95398832257958, 0.3848), (7, 0.9432692889277, 0.2328)]) def test_against_ppcc(self, case): # Test against R ppcc, e.g. # library(ppcc) # options(digits=16) # x < - c(0.52325412, 1.06907699, -0.36084066, 0.15305959, 0.99093194) # set.seed(100) # ppccTest(x, "qrayleigh", ppos="Filliben") n, ref_statistic, ref_pvalue = case rng = np.random.default_rng(7777775561439803116) x = rng.normal(size=n) res = stats.goodness_of_fit(stats.rayleigh, x, statistic="filliben", random_state=rng) assert_allclose(res.statistic, ref_statistic, rtol=1e-4) assert_allclose(res.pvalue, ref_pvalue, atol=1.5e-2) def test_params_effects(self): # Ensure that `guessed_params`, `fit_params`, and `known_params` have # the intended effects. rng = np.random.default_rng(9121950977643805391) x = stats.skewnorm.rvs(-5.044559778383153, loc=1, scale=2, size=50, random_state=rng) # Show that `guessed_params` don't fit to the guess, # but `fit_params` and `known_params` respect the provided fit guessed_params = {'c': 13.4} fit_params = {'scale': 13.73} known_params = {'loc': -13.85} rng = np.random.default_rng(9121950977643805391) res1 = goodness_of_fit(stats.weibull_min, x, n_mc_samples=2, guessed_params=guessed_params, fit_params=fit_params, known_params=known_params, random_state=rng) assert not np.allclose(res1.fit_result.params.c, 13.4) assert_equal(res1.fit_result.params.scale, 13.73) assert_equal(res1.fit_result.params.loc, -13.85) # Show that changing the guess changes the parameter that gets fit, # and it changes the null distribution guessed_params = {'c': 2} rng = np.random.default_rng(9121950977643805391) res2 = goodness_of_fit(stats.weibull_min, x, n_mc_samples=2, guessed_params=guessed_params, fit_params=fit_params, known_params=known_params, random_state=rng) assert not np.allclose(res2.fit_result.params.c, res1.fit_result.params.c, rtol=1e-8) assert not np.allclose(res2.null_distribution, res1.null_distribution, rtol=1e-8) assert_equal(res2.fit_result.params.scale, 13.73) assert_equal(res2.fit_result.params.loc, -13.85) # If we set all parameters as fit_params and known_params, # they're all fixed to those values, but the null distribution # varies. fit_params = {'c': 13.4, 'scale': 13.73} rng = np.random.default_rng(9121950977643805391) res3 = goodness_of_fit(stats.weibull_min, x, n_mc_samples=2, guessed_params=guessed_params, fit_params=fit_params, known_params=known_params, random_state=rng) assert_equal(res3.fit_result.params.c, 13.4) assert_equal(res3.fit_result.params.scale, 13.73) assert_equal(res3.fit_result.params.loc, -13.85) assert not np.allclose(res3.null_distribution, res1.null_distribution) def test_custom_statistic(self): # Test support for custom statistic function. # References: # [1] Pyke, R. (1965). "Spacings". Journal of the Royal Statistical # Society: Series B (Methodological), 27(3): 395-436. # [2] Burrows, P. M. (1979). "Selected Percentage Points of # Greenwood's Statistics". Journal of the Royal Statistical # Society. Series A (General), 142(2): 256-258. # Use the Greenwood statistic for illustration; see [1, p.402]. def greenwood(dist, data, *, axis): x = np.sort(data, axis=axis) y = dist.cdf(x) d = np.diff(y, axis=axis, prepend=0, append=1) return np.sum(d ** 2, axis=axis) # Run the Monte Carlo test with sample size = 5 on a fully specified # null distribution, and compare the simulated quantiles to the exact # ones given in [2, Table 1, column (n = 5)]. rng = np.random.default_rng(9121950977643805391) data = stats.expon.rvs(size=5, random_state=rng) result = goodness_of_fit(stats.expon, data, known_params={'loc': 0, 'scale': 1}, statistic=greenwood, random_state=rng) p = [.01, .05, .1, .2, .3, .4, .5, .6, .7, .8, .9, .95, .99] exact_quantiles = [ .183863, .199403, .210088, .226040, .239947, .253677, .268422, .285293, .306002, .334447, .382972, .432049, .547468] simulated_quantiles = np.quantile(result.null_distribution, p) assert_allclose(simulated_quantiles, exact_quantiles, atol=0.005) class TestFitResult: def test_plot_iv(self): rng = np.random.default_rng(1769658657308472721) data = stats.norm.rvs(0, 1, size=100, random_state=rng) def optimizer(*args, **kwargs): return differential_evolution(*args, **kwargs, seed=rng) bounds = [(0, 30), (0, 1)] res = stats.fit(stats.norm, data, bounds, optimizer=optimizer) try: import matplotlib # noqa: F401 message = r"`plot_type` must be one of \{'..." with pytest.raises(ValueError, match=message): res.plot(plot_type='llama') except (ModuleNotFoundError, ImportError): # Avoid trying to call MPL with numpy 2.0-dev, because that fails # too often due to ABI mismatches and is hard to avoid. This test # will work fine again once MPL has done a 2.0-compatible release. if not np.__version__.startswith('2.0.0.dev0'): message = r"matplotlib must be installed to use method `plot`." with pytest.raises(ModuleNotFoundError, match=message): res.plot(plot_type='llama')