File: test_bootstrap.py

package info (click to toggle)
python-resample 1.10.1-2
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 1,228 kB
  • sloc: python: 1,542; makefile: 12
file content (507 lines) | stat: -rw-r--r-- 14,421 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
# ruff: noqa: D100 D103
import numpy as np
import pytest
from numpy.testing import assert_equal, assert_allclose
from scipy import stats

from resample.bootstrap import (
    _fit_parametric_family,
    bootstrap,
    confidence_interval,
    resample,
    variance,
    covariance,
)

PARAMETRIC_CONTINUOUS = {
    # use scipy.stats names here
    "norm",
    "t",
    "laplace",
    "logistic",
    "f",
    "beta",
    "gamma",
    "lognorm",
    "invgauss",
    "pareto",
}
PARAMETRIC_DISCRETE = {"poisson"}
PARAMETRIC = PARAMETRIC_CONTINUOUS | PARAMETRIC_DISCRETE
NON_PARAMETRIC = {"ordinary", "balanced"}
ALL_METHODS = NON_PARAMETRIC | PARAMETRIC


def chisquare(
    obs, exp=None
):  # we do not use scipy.stats.chisquare, because it is broken
    n = len(obs)
    if exp is None:
        exp = 1.0 / n
    t = np.sum(obs**2 / exp) - n
    return stats.chi2(n - 1).cdf(t)


@pytest.fixture
def rng():
    return np.random.Generator(np.random.PCG64(1))


@pytest.mark.parametrize("method", ALL_METHODS)
def test_resample_shape_1d(method):
    if method == "beta":
        x = (0.1, 0.2, 0.3)
    else:
        x = (1.0, 2.0, 3.0)
    n_rep = 5
    count = 0
    with np.errstate(invalid="ignore"):
        for bx in resample(x, size=n_rep, method=method):
            assert len(bx) == len(x)
            count += 1
    assert count == n_rep


@pytest.mark.parametrize("method", NON_PARAMETRIC | {"norm"})
def test_resample_shape_2d(method):
    x = [(1.0, 2.0), (4.0, 3.0), (6.0, 5.0)]
    n_rep = 5
    count = 0
    for bx in resample(x, size=n_rep, method=method):
        assert bx.shape == np.shape(x)
        count += 1
    assert count == n_rep


@pytest.mark.parametrize("method", NON_PARAMETRIC)
def test_resample_shape_4d(method):
    x = np.ones((2, 3, 4, 5))
    n_rep = 5
    count = 0
    for bx in resample(x, size=n_rep, method=method):
        assert bx.shape == np.shape(x)
        count += 1
    assert count == n_rep


@pytest.mark.parametrize("method", NON_PARAMETRIC | PARAMETRIC_CONTINUOUS)
def test_resample_1d_statistical_test(method, rng):
    # distribution parameters for parametric families
    args = {
        "t": (2,),
        "f": (25, 20),
        "beta": (2, 1),
        "gamma": (1.5,),
        "lognorm": (1.0,),
        "invgauss": (1,),
        "pareto": (1,),
    }.get(method, ())

    if method in NON_PARAMETRIC:
        dist = stats.norm
    else:
        dist = getattr(stats, method)

    x = dist.rvs(*args, size=1000, random_state=rng)

    # make equidistant bins in quantile space for this particular data set
    with np.errstate(invalid="ignore"):
        par = _fit_parametric_family(dist, x)
    prob = np.linspace(0, 1, 11)
    xe = dist(*par).ppf(prob)

    # - in case of parametric bootstrap, wref is exactly uniform
    # - in case of ordinary and balanced, it needs to be computed from original sample
    if method in NON_PARAMETRIC:
        wref = np.histogram(x, bins=xe)[0]
    else:
        wref = len(x) / (len(xe) - 1)

    # compute P values for replicates compared to original
    prob = []
    wsum = 0
    with np.errstate(invalid="ignore"):
        for bx in resample(x, size=100, method=method, random_state=rng):
            w = np.histogram(bx, bins=xe)[0]
            wsum += w
            pvalue = chisquare(w, wref)
            prob.append(pvalue)

    if method == "balanced":
        # balanced bootstrap exactly reproduces frequencies in original sample
        assert_equal(wref * 100, wsum)

    # check whether P value distribution is flat
    # - test has chance probability of 1 % to fail randomly
    # - if it fails due to programming error, value is typically < 1e-20
    wp = np.histogram(prob, range=(0, 1))[0]
    pvalue = chisquare(wp)
    assert pvalue > 0.01


def test_resample_1d_statistical_test_poisson(rng):
    # poisson is behaving super weird in scipy
    x = rng.poisson(1.5, size=1000)
    mu = np.mean(x)

    xe = (0, 1, 2, 3, 10)
    # somehow location 1 is needed here...
    wref = np.diff(stats.poisson(mu, 1).cdf(xe)) * len(x)

    # compute P values for replicates compared to original
    prob = []
    for bx in resample(x, size=100, method="poisson", random_state=rng):
        w = np.histogram(bx, bins=xe)[0]

        pvalue = chisquare(w, wref)
        prob.append(pvalue)

    # check whether P value distribution is flat
    # - test has chance probability of 1 % to fail randomly
    # - if it fails due to programming error, value is typically < 1e-20
    wp = np.histogram(prob, range=(0, 1))[0]
    pvalue = chisquare(wp)
    assert pvalue > 0.01


def test_resample_invalid_family_raises():
    msg = "Invalid family"
    with pytest.raises(ValueError, match=msg):
        next(resample((1, 2, 3), method="foobar"))


@pytest.mark.parametrize("method", PARAMETRIC - {"norm"})
def test_resample_2d_parametric_raises(method):
    with pytest.raises(ValueError):
        next(resample(np.ones((2, 2)), method=method))


def test_resample_3d_parametric_normal_raises():
    with pytest.raises(ValueError):
        next(resample(np.ones((2, 2, 2)), method="normal"))


def test_resample_equal_along_axis():
    data = np.reshape(np.tile([0, 1, 2], 3), (3, 3))
    for b in resample(data, size=2):
        assert_equal(data, b)


@pytest.mark.parametrize("method", NON_PARAMETRIC)
def test_resample_full_strata(method):
    data = np.arange(3)
    for b in resample(data, size=2, strata=data, method=method):
        assert_equal(data, b)


def test_resample_invalid_strata_raises():
    msg = "must have the same shape"
    with pytest.raises(ValueError, match=msg):
        next(resample((1, 2, 3), strata=np.arange(4)))


def test_bootstrap_2d_balanced(rng):
    data = ((1, 2, 3), (2, 3, 4), (3, 4, 5))

    def mean(x):
        return np.mean(x, axis=0)

    r = bootstrap(mean, data, method="balanced")

    # arithmetic mean is linear, therefore mean over all replicates in
    # balanced bootstrap is equal to mean of original sample
    assert_allclose(mean(data), mean(r))


@pytest.mark.parametrize("action", [bootstrap, variance, confidence_interval])
def test_bootstrap_several_args(action):
    x = [1, 2, 3]
    y = [4, 5, 6]
    xy = np.transpose([x, y])

    if action is confidence_interval:

        def f1(x, y):
            return np.sum(x + y)

        def f2(xy):
            return np.sum(xy)

    else:

        def f1(x, y):
            return np.sum(x), np.sum(y)

        def f2(xy):
            return np.sum(xy, axis=0)

    r1 = action(f1, x, y, size=10, random_state=1)
    r2 = action(f2, xy, size=10, random_state=1)

    assert_equal(r1, r2)


@pytest.mark.parametrize("ci_method", ["percentile", "bca"])
def test_confidence_interval(ci_method, rng):
    data = rng.normal(size=1000)
    par = stats.norm.fit(data)
    dist = stats.norm(
        par[0], par[1] / len(data) ** 0.5
    )  # accuracy of mean is sqrt(n) better
    cl = 0.9
    ci_ref = dist.ppf(0.05), dist.ppf(0.95)
    ci = confidence_interval(
        np.mean, data, cl=cl, size=1000, ci_method=ci_method, random_state=rng
    )
    assert_allclose(ci_ref, ci, atol=6e-3)


def test_confidence_interval_invalid_p_raises():
    msg = "must be between zero and one"
    with pytest.raises(ValueError, match=msg):
        confidence_interval(np.mean, (1, 2, 3), cl=2)


def test_confidence_interval_invalid_ci_method_raises():
    msg = "method must be 'percentile' or 'bca'"
    with pytest.raises(ValueError, match=msg):
        confidence_interval(np.mean, (1, 2, 3), ci_method="foobar")


def test_bca_confidence_interval_estimator_returns_int(rng):
    def fn(data):
        return int(np.mean(data))

    data = (1, 2, 3)
    ci = confidence_interval(fn, data, ci_method="bca", size=5, random_state=rng)
    assert_allclose((1.0, 2.0), ci)


@pytest.mark.parametrize("ci_method", ["percentile", "bca"])
def test_bca_confidence_interval_bounded_estimator(ci_method, rng):
    def fn(data):
        return max(np.mean(data), 0)

    data = (-3, -2, -1)
    ci = confidence_interval(fn, data, ci_method=ci_method, size=5, random_state=rng)
    assert_allclose((0.0, 0.0), ci)


@pytest.mark.parametrize("method", NON_PARAMETRIC)
def test_variance(method, rng):
    data = np.arange(100)
    v = np.var(data) / len(data)

    r = variance(np.mean, data, size=1000, method=method, random_state=rng)
    assert r == pytest.approx(v, rel=0.05)


@pytest.mark.parametrize("method", NON_PARAMETRIC)
def test_covariance(method, rng):
    cov = np.array([[1.0, 0.1], [0.1, 2.0]])
    data = rng.multivariate_normal([0.1, 0.2], cov, size=1000)

    r = covariance(
        lambda x: np.mean(x, axis=0), data, size=1000, method=method, random_state=rng
    )
    assert_allclose(r, cov / len(data), atol=1e-3)


def test_resample_deprecation(rng):
    data = [1, 2, 3]

    with pytest.warns(FutureWarning):
        r = list(resample(data, 10))
        assert np.shape(r) == (10, 3)

    with pytest.warns(FutureWarning):
        resample(data, 10, "balanced")

    with pytest.warns(FutureWarning):
        with pytest.raises(ValueError):
            resample(data, 10, "foo")

    with pytest.warns(FutureWarning):
        resample(data, 10, "balanced", [1, 1, 2])

    with pytest.warns(FutureWarning):
        with pytest.raises(ValueError):
            resample(data, 10, "balanced", [1, 1])

    with pytest.warns(FutureWarning):
        resample(data, 10, "balanced", [1, 1, 2], rng)

    with pytest.warns(FutureWarning):
        resample(data, 10, "balanced", [1, 1, 2], 1)

    with pytest.warns(FutureWarning):
        with pytest.raises(TypeError):
            resample(data, 10, "balanced", [1, 1, 2], 1.3)

    with pytest.warns(FutureWarning):
        with pytest.raises(ValueError):  # too many arguments
            resample(data, 10, "balanced", [1, 1, 2], 1, 2)


def test_confidence_interval_deprecation(rng):
    d = [1, 2, 3]
    with pytest.warns(FutureWarning):
        r = confidence_interval(np.mean, d, 0.6, random_state=1)
    assert_equal(r, confidence_interval(np.mean, d, cl=0.6, random_state=1))

    with pytest.warns(FutureWarning):
        r = confidence_interval(np.mean, d, 0.6, "percentile", random_state=1)
    assert_equal(
        r,
        confidence_interval(np.mean, d, cl=0.6, ci_method="percentile", random_state=1),
    )

    with pytest.warns(FutureWarning):
        with pytest.raises(ValueError):
            confidence_interval(np.mean, d, 0.6, "percentile", 1)


def test_random_state():
    d = [1, 2, 3]
    a = list(resample(d, size=5, random_state=np.random.default_rng(1)))
    b = list(resample(d, size=5, random_state=1))
    c = list(resample(d, size=5, random_state=[2, 3]))
    assert_equal(a, b)
    assert not np.all([np.all(ai == ci) for (ai, ci) in zip(a, c)])

    with pytest.raises(TypeError):
        resample(d, size=5, random_state=1.5)


@pytest.mark.parametrize("method", NON_PARAMETRIC)
def test_resample_several_args(method):
    a = [1, 2, 3]
    b = [(1, 2), (2, 3), (3, 4)]
    c = ["12", "3", "4"]
    r1 = [[], [], []]
    for ai, bi, ci in resample(a, b, c, size=5, method=method, random_state=1):
        r1[0].append(ai)
        r1[1].append(bi)
        r1[2].append(ci)

    r2 = [[], [], []]
    abc = np.empty(3, dtype=[("a", "i"), ("b", "i", 2), ("c", "U4")])
    abc[:]["a"] = a
    abc[:]["b"] = b
    abc[:]["c"] = c
    for abci in resample(abc, size=5, method=method, random_state=1):
        r2[0].append(abci["a"])
        r2[1].append(abci["b"])
        r2[2].append(abci["c"])

    for i in range(3):
        assert_equal(r1[i], r2[i])


def test_resample_several_args_incompatible_keywords():
    a = [1, 2, 3]
    b = [(1, 2), (2, 3), (3, 4)]
    with pytest.raises(ValueError):
        resample(a, b, size=5, method="norm")

    resample(a, size=5, strata=[1, 1, 2])

    with pytest.raises(ValueError):
        resample(a, b, size=5, strata=[1, 1, 2])

    resample(a, b, a, b, size=5)

    with pytest.raises(ValueError):
        resample(a, [1, 2])

    with pytest.raises(ValueError):
        resample(a, [1, 2, 3, 4])

    with pytest.raises(ValueError):
        resample(a, b, 5)


def test_resample_extended_1():
    a = [1, 2, 3]
    bs = list(resample(a, size=100, method="extended", random_state=1))

    # check that lengths of bootstrap samples are poisson distributed
    w, xe = np.histogram([len(b) for b in bs], bins=10, range=(0, 10))
    wm = stats.poisson(len(a)).pmf(xe[:-1]) * np.sum(w)
    t = np.sum((w - wm) ** 2 / wm)
    pvalue = 1 - stats.chi2(len(w)).cdf(t)
    assert pvalue > 0.1


def test_resample_extended_2():
    n = 10
    a = np.arange(n)
    ts = []
    for b in resample(a, size=1000, method="extended", random_state=1):
        ts.append(np.mean(b))

    t = np.var(ts)
    expected_not_extended = np.var(a) / n

    k = np.arange(100)
    pk = stats.poisson(n).pmf(k)
    expected = expected_not_extended * np.sum(pk[1:] * n / k[1:]) / (1 - pk[0])

    assert expected / expected_not_extended > 1.1
    assert t > expected_not_extended
    assert_allclose(t, expected, atol=0.02)


def test_resample_extended_3():
    n = 10
    a = np.arange(n)
    b = 5 + a
    ns = []
    for ai, bi in resample(a, b, size=1000, method="extended", random_state=1):
        assert len(ai) == len(bi)
        assert_equal(bi - ai, 5)
        ns.append(len(ai))
    assert_allclose(np.var(ns), 10, rtol=0.05)


def test_resample_extended_4():
    x = np.ones(10)
    a = np.transpose((x, 3 * x))

    ts = []
    for b in resample(a, size=1000, method="extended", random_state=1):
        ts.append(np.sum(b, axis=0))

    t = np.var(ts, axis=0)

    mu = np.sum(x, axis=0)
    assert_allclose(t, (mu, 3**2 * mu), rtol=0.05)


def test_resample_extended_5():
    x = np.ones(10)
    a = np.transpose((x, 3 * x))

    ts1 = []
    ts2 = []
    for b1, b2 in resample(a, 3 * a, size=1000, method="extended", random_state=1):
        ts1.append(np.sum(b1, axis=0))
        ts2.append(np.sum(b2, axis=0))

    t1 = np.var(ts1, axis=0)
    t2 = np.var(ts2, axis=0)

    mu1 = np.sum(x, axis=0)
    mu2 = 3**2 * np.sum(x, axis=0)
    assert_allclose(t1, (mu1, 3**2 * mu1), rtol=0.05)
    assert_allclose(t2, (mu2, 3**2 * mu2), rtol=0.05)


def test_bias_error():
    with pytest.raises(NotImplementedError):
        from resample.bootstrap import bias  # noqa

    with pytest.raises(NotImplementedError):
        import resample.bootstrap as b

        b.bias_corrected  # noqa