ENH Replaced RandomState with Generator compatible calls #22271
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
RandomState.random_sample calls to RandomState.uniformRandomState.random_sample calls to RandomState.uniform
RandomState.random_sample calls to RandomState.uniformRandomState.random_sample calls to RandomState.uniform
|
There looks to be a few more for grep -r "random_sample" sklearn --exclude="*tests*" --exclude="*.pyc" \
--exclude="*.pxi" --exclude="*.c" --exclude="*.html" --exclude="*.so" -inResultssklearn/ensemble/_gb.py:36:from ._gradient_boosting import _random_sample_mask
sklearn/ensemble/_gb.py:660: sample_mask = _random_sample_mask(n_samples, n_inbag, random_state)
sklearn/ensemble/_gradient_boosting.pyx:237:def _random_sample_mask(np.npy_intp n_total_samples,
sklearn/cluster/_kmeans.py:212: rand_vals = random_state.random_sample(n_local_trials) * current_pot
sklearn/multiclass.py:1058: self.code_book_ = random_state.random_sample((n_classes, code_size_))
sklearn/neural_network/_rbm.py:199: return rng.random_sample(size=p.shape) < p
sklearn/neural_network/_rbm.py:220: return rng.random_sample(size=p.shape) < p
sklearn/neighbors/_kde.py:108: >>> X = rng.random_sample((100, 3))For grep -r "randn" sklearn --exclude="*tests*" --exclude="*.pyc" \
--exclude="*.pxi" --exclude="*.c" --exclude="*.html" --exclude="*.so" -inResultssklearn/cluster/_affinity_propagation.py:174: ) * random_state.randn(n_samples, n_samples)
sklearn/datasets/_samples_generator.py:233: X[:, :n_informative] = generator.randn(n_samples, n_informative)
sklearn/datasets/_samples_generator.py:262: X[:, -n_useless:] = generator.randn(n_samples, n_useless)
sklearn/datasets/_samples_generator.py:598: X = generator.randn(n_samples, n_features)
sklearn/datasets/_samples_generator.py:1025: + noise * generator.randn(n_samples)
sklearn/datasets/_samples_generator.py:1091: ) ** 0.5 + noise * generator.randn(n_samples)
sklearn/datasets/_samples_generator.py:1156: ) + noise * generator.randn(n_samples)
sklearn/datasets/_samples_generator.py:1221: u, _ = linalg.qr(generator.randn(n_samples, n), mode="economic", check_finite=False)
sklearn/datasets/_samples_generator.py:1223: generator.randn(n_features, n), mode="economic", check_finite=False
sklearn/datasets/_samples_generator.py:1283: D = generator.randn(n_features, n_components)
sklearn/datasets/_samples_generator.py:1292: X[idx, i] = generator.randn(n_nonzero_coefs)
sklearn/datasets/_samples_generator.py:1522: X += noise * generator.randn(3, n_samples)
sklearn/datasets/_samples_generator.py:1564: X += noise * generator.randn(3, n_samples)
sklearn/linear_model/_quantile.py:92: >>> y = rng.randn(n_samples)
sklearn/linear_model/_quantile.py:93: >>> X = rng.randn(n_samples, n_features)
sklearn/linear_model/_sag.py:216: >>> X = rng.randn(n_samples, n_features)
sklearn/linear_model/_sag.py:217: >>> y = rng.randn(n_samples)
sklearn/linear_model/_ridge.py:971: >>> y = rng.randn(n_samples)
sklearn/linear_model/_ridge.py:972: >>> X = rng.randn(n_samples, n_features)
sklearn/linear_model/_stochastic_gradient.py:1890: >>> y = rng.randn(n_samples)
sklearn/linear_model/_stochastic_gradient.py:1891: >>> X = rng.randn(n_samples, n_features)
sklearn/kernel_ridge.py:126: >>> y = rng.randn(n_samples)
sklearn/kernel_ridge.py:127: >>> X = rng.randn(n_samples, n_features)
sklearn/utils/estimator_checks.py:171: X = rng.randn(10, 5)
sklearn/feature_selection/_mutual_info.py:293: 1e-10 * means * rng.randn(n_samples, np.sum(continuous_mask))
sklearn/feature_selection/_mutual_info.py:298: y += 1e-10 * np.maximum(1, np.mean(np.abs(y))) * rng.randn(n_samples)
sklearn/svm/_classes.py:1196: >>> y = rng.randn(n_samples)
sklearn/svm/_classes.py:1197: >>> X = rng.randn(n_samples, n_features)
sklearn/svm/_classes.py:1388: >>> y = np.random.randn(n_samples)
sklearn/svm/_classes.py:1389: >>> X = np.random.randn(n_samples, n_features)
sklearn/manifold/_t_sne.py:994: X_embedded = 1e-4 * random_state.randn(n_samples, self.n_components).astype(
sklearn/manifold/_spectral_embedding.py:339: X = random_state.randn(laplacian.shape[0], n_components + 1)
sklearn/manifold/_spectral_embedding.py:370: X = random_state.randn(laplacian.shape[0], n_components + 1)
sklearn/mixture/_base.py:459: mean + rng.randn(sample, n_features) * np.sqrt(covariance)
sklearn/model_selection/_split.py:1006: >>> X = np.random.randn(12, 2)
sklearn/decomposition/_nmf.py:317: H = avg * rng.randn(n_components, n_features).astype(X.dtype, copy=False)
sklearn/decomposition/_nmf.py:318: W = avg * rng.randn(n_samples, n_components).astype(X.dtype, copy=False)
sklearn/decomposition/_nmf.py:372: W[W == 0] = abs(avg * rng.randn(len(W[W == 0])) / 100)
sklearn/decomposition/_nmf.py:373: H[H == 0] = abs(avg * rng.randn(len(H[H == 0])) / 100)
sklearn/neighbors/_nca.py:454: transformation = self.random_state_.randn(n_components, X.shape[1]) |
|
Thank you for that. I'm on windows and their string searching/parsing utilities aren't...great. Will update the PR |
|
How would I log this in the changelog? |
I do not think this needs a change log since there should be changes to the user. Title should be updated tho since the scope increased. (Test failing is unrelated, looking into it now) |
RandomState.random_sample calls to RandomState.uniformRandomState-specific calls to equivalent calls that match signature with Generator calls
RandomState-specific calls to equivalent calls that match signature with Generator callsRandomState-specific calls to equivalent calls that match signature with Generator calls
RandomState-specific calls to equivalent calls that match signature with Generator callsRandomState-specific calls to equivalent calls that match signature with Generator calls
thomasjpfan
reviewed
Jan 25, 2022
There was a problem hiding this comment.
LGTM!
For future reviewers, this change does not change any generated random numbers. It's more so to make NumPy Generators easier to adopt.
-
For reference,
randn->standard_normalare the same call: https://github.com/numpy/numpy/blob/6077afd650a503034d0a8a5917bb9a5fa3f115fd/numpy/random/mtrand.pyx#L1243-L1246 -
randcallsrandom_sample: https://github.com/numpy/numpy/blob/6077afd650a503034d0a8a5917bb9a5fa3f115fd/numpy/random/mtrand.pyx#L1179-L1182 -
random_sample->uniformgenerate the same values as they have use the same underlyingCcode. Through testing, they are all the same:
import numpy as np
from numpy.testing import assert_allclose
for i in range(20):
rng1 = np.random.RandomState(i)
rng2 = np.random.RandomState(i)
for row, col in zip(range(0, 1000, 100), range(0, 1000, 100)):
x1 = rng1.random_sample((row, col))
x2 = rng2.uniform(size=(row, col))
assert_allclose(x1, x2)The only difference is that uniform expands it a little by allowing for low and high.
RandomState-specific calls to equivalent calls that match signature with Generator calls|
As a follow up PR, there are grep -r "rand(" sklearn --exclude="*tests*" --exclude="*.pyc" --exclude="*.cpp" --exclude="*.h" \
--exclude="*.pxi" --exclude="*.c" --exclude="*.html" --exclude="*.so" -inResults (Could be some false positives)sklearn/metrics/pairwise.py:1655: >>> X = np.random.RandomState(0).rand(5, 3)
sklearn/ensemble/_gradient_boosting.pyx:259: random_state.rand(n_total_samples)
sklearn/semi_supervised/_label_propagation.py:39:>>> random_unlabeled_points = rng.rand(len(iris.target)) < 0.3
sklearn/semi_supervised/_label_propagation.py:411: >>> random_unlabeled_points = rng.rand(len(iris.target)) < 0.3
sklearn/semi_supervised/_label_propagation.py:567: >>> random_unlabeled_points = rng.rand(len(iris.target)) < 0.3
sklearn/semi_supervised/_self_training.py:133: >>> random_unlabeled_points = rng.rand(iris.target.shape[0]) < 0.3
sklearn/datasets/_samples_generator.py:229: centroids *= generator.rand(n_clusters, 1)
sklearn/datasets/_samples_generator.py:230: centroids *= generator.rand(1, n_informative)
sklearn/datasets/_samples_generator.py:242: A = 2 * generator.rand(n_informative, n_informative) - 1
sklearn/datasets/_samples_generator.py:249: B = 2 * generator.rand(n_informative, n_redundant) - 1
sklearn/datasets/_samples_generator.py:257: indices = ((n - 1) * generator.rand(n_repeated) + 0.5).astype(np.intp)
sklearn/datasets/_samples_generator.py:266: flip_mask = generator.rand(n_samples) < flip_y
sklearn/datasets/_samples_generator.py:271: shift = (2 * generator.rand(n_features) - 1) * class_sep
sklearn/datasets/_samples_generator.py:275: scale = 1 + 100 * generator.rand(n_features)
sklearn/datasets/_samples_generator.py:394: p_c = generator.rand(n_classes)
sklearn/datasets/_samples_generator.py:397: p_w_c = generator.rand(n_features, n_classes)
sklearn/datasets/_samples_generator.py:412: c = np.searchsorted(cumulative_p_c, generator.rand(y_size - len(y)))
sklearn/datasets/_samples_generator.py:430: words = np.searchsorted(cumulative_p_w_sample, generator.rand(n_words))
sklearn/datasets/_samples_generator.py:614: ground_truth[:n_informative, :] = 100 * generator.rand(n_informative, n_targets)
sklearn/datasets/_samples_generator.py:1019: X = generator.rand(n_samples, n_features)
sklearn/datasets/_samples_generator.py:1082: X = generator.rand(n_samples, 4)
sklearn/datasets/_samples_generator.py:1147: X = generator.rand(n_samples, 4)
sklearn/datasets/_samples_generator.py:1377: A = generator.rand(n_dim, n_dim)
sklearn/datasets/_samples_generator.py:1379: X = np.dot(np.dot(U, 1.0 + np.diag(generator.rand(n_dim))), Vt)
sklearn/datasets/_samples_generator.py:1439: aux = random_state.rand(dim, dim)
sklearn/datasets/_samples_generator.py:1443: ) * random_state.rand(np.sum(aux > alpha))
sklearn/datasets/_samples_generator.py:1507: t = 1.5 * np.pi * (1 + 2 * generator.rand(n_samples))
sklearn/datasets/_samples_generator.py:1508: y = 21 * generator.rand(n_samples)
sklearn/datasets/_samples_generator.py:1515: parameters = generator.rand(2, n_samples) * np.array([[np.pi], [7]])
sklearn/datasets/_samples_generator.py:1558: t = 3 * np.pi * (generator.rand(1, n_samples) - 0.5)
sklearn/datasets/_samples_generator.py:1560: y = 2.0 * generator.rand(1, n_samples)
sklearn/random_projection.py:512: >>> X = rng.rand(25, 3000)
sklearn/random_projection.py:662: >>> X = rng.rand(25, 3000)
sklearn/utils/estimator_checks.py:801: X = rng.rand(40, 3)
sklearn/utils/estimator_checks.py:805: y = (4 * rng.rand(40)).astype(int)
sklearn/utils/estimator_checks.py:1091: X = _pairwise_estimator_convert_X(rng.rand(40, 10), estimator_orig)
sklearn/utils/random.py:92: class_probability_nz_norm.cumsum(), rng.rand(nnz)
sklearn/manifold/_mds.py:87: X = random_state.rand(n_samples * n_components)
sklearn/mixture/_base.py:151: resp = random_state.rand(n_samples, self.n_components)
sklearn/decomposition/_truncated_svd.py:143: >>> X_dense = np.random.rand(100, 100) |
|
This is a great first step for Generator, thanks! |
7 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Reference Issues/PRs
Issue #20669
Towards to #16988
What does this implement/fix? Explain your changes.
Per #20669 it is discussed that we will likely need an eventual change towards adopting NumPy's
Generatorinterface instead of remaining onRandomState. To ease such a transition, this PR replacesRandomState.random_samplecalls with their correspondingRandomState.uniformto match theGenerator.uniformsyntax, preserving functionality but allowing for a drop-in replacement of the underlying object without syntax errors.Any other comments?
Thank you to @thomasjpfan for the direction and guidance for this PR.
In working on this PR, I also looked similar changing RandomState methods to equivalent methods w/ overlapping names/signatures between the two interfaces, namely:
randn -> standard_normal,random_sample, rand -> uniform.