Add statistical test for regex-guided generation #77
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
… simple regex. It has a frozen seed to avoid flakiness, but there is information in there to make it a proper probabilistic test. It's also possible to reduce the run length by taking fewer samples.
brandonwillard
requested changes
Oct 17, 2024
… the computation of the exact distributions easier. Tests of the variance and a Kolmogorov-Smirnov test for the distribution has been added. Asserts have been computed for n=250. If we want the test to run faster, we can reduce this number. All tests have been run with n=1000000 to verify the monte carlo error goes to zero. All tests have been run with multiple seeds.
There was a problem hiding this comment.
It looks like we could generalize some of this code and move it outside of the test function. That would make it easier to separate the Markov and non-Markov tests so that they could be run separately/in parallel (e.g. locally you can use pytest-xdist and fly through the tests without changes to the code).
This isn't necessary, though, so no worries.
Other than that, we can squash and merge after the above two edits.
|
I agree this can be generalized. Will save that work for when the next version is added though. |
|
@brandonwillard Foolishly made the changes myself rather than accepting yours, so I can't merge. Sorry. |
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Outdated suggestions cannot be applied.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.
I added a simple test that checks that average length of a regex-guided generation on a simple grammar with a simple language model and a fairly simple regex described in #73 is correct. For simplicity, this has a fixed seed, however there is enough information in the issue to turn it into a stochastic test is someone has the stomach for that.
It would be possible to make this test run faster with fewer Monte Carlo samples. I have verified it on much larger sample sizes and it's consistently within MC-standard error of the correct solution. I have also varied parameters in the language model.
I had to kludge together a
generatefunction so as to not depend onoutlines, but I think I'm using the guides properly.