Filter positive items for ranking evaluation #523
Merged
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
|
Hi @amirj, we're making some changes to how the ranking metrics being computed, and these changes will break the logic of PropensityStratifiedEvaluation. Could you please have a look and see if we could mitigate this with some modification to the eval method? |
saghiles
reviewed
Aug 10, 2023
saghiles
reviewed
Aug 10, 2023
tqtg
force-pushed
the
filter-positive-items-for-ranking-eval
branch
from
6c5f332 to
3a1f0ce
Compare
This was referenced May 7, 2024
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Outdated suggestions cannot be applied.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.
Description
Previously, we did not filter positive items in training when doing evaluation on test/validation set for most of the ranking metrics. As a result, it harms performance of some models that indeed rank those items to the top. In this PR, we first explicitly filter out those items when calling
Recommender.rank()function usingitem_indices, we then modify all ranking metrics to obey this logic. To summarize, positive items in training will not be involved during evaluation.How are the ranking metrics affected?
AUC: not affected - previous logic already took it into account.Precision: affected - expected the results to be increased.Recall: affected - expected the results to be increased.NDCG: affected - expected the results to be increased.NCRR: affected - expected the results to be increased.MAP: affected - expected the results to be increased.MRR: affected - expected the results to be increased.How are the rating metrics affected?
No. This change only affects ranking metrics.
Related Issues
#503
Checklist:
README.md(if you are adding a new model).examples/README.md(if you are adding a new example).datasets/README.md(if you are adding a new dataset).