Unbiased PopulationSimpleSliceSampler#178
Open
njzifjoiez wants to merge 7 commits into
Open
Conversation
added 6 commits
January 22, 2026 12:08
Merging with the last master before merge request
Owner
|
Nice work! After some more thought, I am no longer surprised that it is biased (I will write about that in the issue), but it seems like it is still an improvement. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hi there,
I finally found some time to look at #170. The code modifications are quite small. I only changed
slice_limit_to_scaleto include random slice placement. Using the bounds of the unit cube as the slice limit should be unbiased too, it is one of the methods proposed in Neal's paper, so I kept it that way.I end up doing a couple of tests with a Rosenbrock 30D with the normal slice sampler, this slice sampler with the slice to the unit cube and finally with the scale. I use the scale adaptation, but not the shrink factor or jitter. With the modification, it appears to be slightly biased depending on how the scale is adapted. From my tests, it seems to perform better at recovering the same logz as the others with stronger adaptation, like 0.8 or 0.7. In all cases, it leads to fewer likelihood calls than the others.
Here is the normal slice sampler as a reference:
Here are some tests with popsize=1:
Same here, but with popsize=10:
Here are the different runs I used to test the factor for scale adaptation. I tested with
scale_adapt_factor=[0.99,0.95,0.85,0.8,0.7,0.5],0.9is the previous test. For 0.8, 0.7 and 0.5, I got log z ~ 173 instead of the ~170 that I got for the less aggressive adaptation.