FeatureHasher should have an option to not hash the values #309
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
oracle-contributor-agreement
bot
added
the
OCA Verified
jhalexand
previously approved these changes
Dec 21, 2022
There was a problem hiding this comment.
Looks good, with the minor nit that you document/describe (accurately) the value hash seed using the phrasing "value hash" but name the member variables and static final members using "hash value". I think the former is clearer as "hash value" can easily be read as a more generic term. I'm still approving since this is a very minor nit, but I'll not merge in case you want to change this first.
…ng validation for dimension.
Craigacp
added a commit
that referenced
this pull request
Dec 21, 2022
* FeatureHasher should have an option to not hash the values. * Renaming DEFAULT_HASH_VALUE_SEED to DEFAULT_VALUE_HASH_SEED, and adding validation for dimension.
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Outdated suggestions cannot be applied.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
FeatureHashernow has an option to preserve the values of the features that it hashes, and this option is available throughTokenPipeline. Adds a couple of tests for the new behaviour.This changes the default behaviour of
TokenPipelinewith hashing, and adds an option to the constructor to revert to the old behaviour. The old behaviour didn't make much sense, but this allows people to preserve their computations. Note that reproducing old models without this fix will turn on the new behaviour as it is defaulted to on, and so must be overridden when reproducing those models.Motivation
A term counting
TokenPipelineshould never return negative values for the term counts, but asFeatureHasherhashes the feature values into{-1,1}this could cause the hashed dimensions to aggregate to a negative value which causes problems elsewhere and semantically isn't term counting.Fixes #307.