Skip to content

Release Notes

Release Notes

The majority of this release comprises new capabilities for federated machine learning.

New Features

  • Support for FATE model training for horizontally federated training data. Supported models types include:
    • federated neural networks (binary & multi-class classification tested; convolution un-tested);
    • federated secureboost decision trees (binary, multi-class, and regression);
    • federated Logistic Regression (binary).
  • Support for data transformation for horizontal federated data. Supported operations include:
    • where;
    • query/filter;
    • fillna;
    • dropna.
  • Support for tabular numeric data in federated modeling.
  • Collection of evaluation metrics (and plots) for FATE models.
  • To support federated machine learning, much of the high-level rest functionality now operates in a federated manner. The features impacted include:
    • Queries
    • EDA trees
    • Custom Ops
    • VDS objects
    • Model definitions
    • FATE Training objects
  • A federation API has been introduced, which allows for the creation, modification, deletion, and retrieval of federate nodes.

Updates and Fixes

  • Gensim updated to 4.0.1
  • Accumulo library migrated to official NSA python package

Known issues

  • FATE model trainings can fail with pika exceptions or rabbitmq exceptions; these are networking issues. Retrying the job will usually work.
    • Often you will see NOT FOUND - no queue
  • FATE model role / party / and initiator settings must be configured properly or the job will fail. Initiator must be the local federate.
  • To use horizontal federated data transformation, the original data query must be filtered to use only horizontal data.