Repository navigation
Reduce allocations and logarithm calculations in categorical TPE - #273
Merged
Merged
Conversation
Categorical TPE allocated an array of category probabilities for each observation. Candidate scoring then recalculated logarithms for every observation. Each observation has one probability for its observed category and another for all other categories. Store the observed category indices and these shared probabilities instead of the arrays. Compute the two logarithms once per density call and reuse them throughout the observation loop. Preserve smoothing, the uniform prior, mixture weights, and correlations between parameters. Add regression tests for probability values, weighted joint densities, and correlated sampling.
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Represent categorical kernels using observed category indices and shared matching/nonmatching probabilities instead of per-observation probability arrays. Compute logarithms outside the observation loop to reduce allocations and repeated calculations during candidate scoring.
Closes #267.
Original implementation by @tkng, cherry-picked from 453a2a2 without code changes, preserving the original commit author.