Skip to content

Feature Quantization [3/5]: Preprocessor (write-side) Updates - #735

Draft
jchmura-sc wants to merge 4 commits into
jchmura/feature_quant_opsfrom
jchmura/feature_quant_preprocessor
Draft

Feature Quantization [3/5]: Preprocessor (write-side) Updates#735
jchmura-sc wants to merge 4 commits into
jchmura/feature_quant_opsfrom
jchmura/feature_quant_preprocessor

Conversation

@jchmura-sc

@jchmura-sc jchmura-sc commented Aug 3, 2026

Copy link
Copy Markdown
Collaborator

Purpose
The purpose of this PR is to make the (write-side) preprocessor changes for feature quantization:

  • FeatureQuantizationSpec the API endpoint that the user can specific inside NodeDataPreprocessingSpec
  • beam transform for computing quantization metadata and serializing to temporary json
  • beam transform for quantizing node features based on that metadata (calls into quantize_ndarray)
  • driver reads the json, constructs FeatureQuantizationMetadata proto, and adds to preprocessed_metadata_pb

Context

Updated Changelog.md? NO

Ready for code review?: NO

return quantized_schema


def _build_feature_values(

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

TBD: Confirming we impact of switching to asymetric clip values. If we proceed then this can be removed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant