Better support string values for scores #785
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
If a user produces a score whose value is a string, when that value is ‘reduced’ using the default mean reducer, it is coerced to a float. For strings thing means when the Score arrives at the custom metric, it will carry the reduced value which has been coerced to a float. This fix is minimal - it implements support for string values in the mean reducer, providing the most common string value (or the first string value if non are most common). Fixes #775
jjallaire
approved these changes
Oct 31, 2024
dragonstyle
added a commit
that referenced
this pull request
Nov 4, 2024
This reverts commit 79f1d7e. # Conflicts: # CHANGELOG.md
jjallaire-aisi
pushed a commit
that referenced
this pull request
Nov 4, 2024
jjallaire-aisi
added a commit
that referenced
this pull request
Nov 4, 2024
* Revert "Better support string values for scores (#785)" This reverts commit 79f1d7e. # Conflicts: # CHANGELOG.md * Add test * Add test * Don’t run reducer against single epoch samples At the lowest level (just before reducing), determine whether there is more than one sample for the sample_id and do not actually reduce the value if there is only one. This prevents side effects of reducers from happening to single samples (side effects include: mutating the scores - strings become floats, for example, due to the mean reducer being applied; mutating the other score fields like explanation and metadata to reduced values; etc). This overall seems good (no side effects when scoring single epoch evaluations), but could introduce bugs if anyone was relying upon those mutations (for example if they wrote string score values, but were downstream expecting those to be forced to floats by the reducer). --------- Co-authored-by: jjallaire-aisi <[email protected]>
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.
If a user produces a score whose value is a string, when that value is ‘reduced’ using the default mean reducer, it is coerced to a float. For strings thing means when the Score arrives at the custom metric, it will carry the reduced value which has been coerced to a float.
This fix is minimal - it implements support for string values in the mean reducer, providing the most common string value (or the first string value if non are most common).
Fixes #775
This PR contains:
What is the current behavior? (You can also link to an open issue here)
Scorers that return string values run through the reduction process and the string values are reduced to a float by the mean reducer.
What is the new behavior?
The mean reducer will reduce strings to the most common value, meaning that the string value is preserved across the mean reducer.
Does this PR introduce a breaking change? (What changes might users need to make in their application due to this PR?)
No
Other information: