IBMFL 1.0.2
Release 1.0.2
IBMFL 1.0.2 is a minor release with improvements and new features such as setting up aggregation quorum, allowing party rejoin and enabling keras training with GPU.
Changes
- Improved Logging by providing better messages.
- TF version requirement changed to
1.15.0
- Sklearn version requirement changed to
0.23.1
. - No API changes.
- No breaking changes.
New functionality
- Quorum: specify quorum percentage and maximum timeout in the aggregator config file to provide flexibility to parties that have potential connectivity failure.
- Rejoin: dropped parties can rejoin training process.
- GPU support: Enabling training with GPU(s) for
KerasFLModel
(you can check our new tutorial).
Bug fixes
- Fix possible feature dimension mismatch for Sklearn multi-class classification models training on Non-iid party distribution.
- Fix save/load inconsistency caused by
pickle
andjoblib
for Sklearn models.