Update RNG usage with numpy's new interface #108
Draft
Conversation
This file contains 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
|
Rebased on the GHA branch used in #109 |
numpy.random.default_rng may use a different algorithm between versions, so testing with a specific algorithm is best for stability. This necessitates updating all of the hard-coded test values. Most are basically the same, but a few are not.
clbarnes
commented
Sep 8, 2022
| assert np.allclose(q, .32697921) | ||
| _, q = bct.modularity_louvain_dir(x, seed=stable_rng) | ||
| # assert np.allclose(q, .32697921) # legacy np.random.RandomState | ||
| assert np.allclose(q, 0.373_475_890) # stable_rng |
There was a problem hiding this comment.
Result with new RNG is quite different to previous.
clbarnes
commented
Sep 8, 2022
| assert np.allclose(q, .07885327) | ||
| # N.B. this result is quite different | ||
| # assert np.allclose(q, .07885327) # legacy numpy.random.RandomState | ||
| assert np.allclose(q, 0.114_732_792) # stable_rng |
There was a problem hiding this comment.
Result with new RNG is quite different to previous.
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Outdated suggestions cannot be applied.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.
get_rngis now just a thin wrapper aroundnumpy.random.default_rng, which reduces the set of types which could be used as a seed but improves consistency.default_rngmay change the underlying algorithm in future numpy releases, which means that results are not necessarily stable for the same seed across numpy versions; as such, users can pass their BitGenerator of choice in theseedargument, and tests are hard-coded to use today's preferred algorithm.Updating from the deprecated RNG (where the algorithm wasn't even selectable) to today's preferred one required changing all of the hardcoded reference values in the tests. For most, this was a pretty minor difference within reason for changing the random seed, but in a couple of cases the change was more significant. I'm not sure how best to check whether this is "real" or not.
Additionally removed EOL python versions from tox and added more recent releases.
Travis may be dead so I'll look into github actions instead.