Secretflow

Added

- Runtime_config as input to utils.testing.cluster_def.
- SFXgboost optimization.
- Horizontal preprocessing
- StardardScaler
- KBinsDiscretizater
- Binning
- Vertical preprocessing: WOE binning and substitution.
- HEU supports int/fxp data type.
- HEU Object supports slice and sum.
- Differential privacy.
- API docstrings.
- Custom pytree node to ppu.
- English docs.

Changed

- Remove default reveal in aggreagtor and compartor.
- Bump dependencies
- jax to 0.3.7
- sf-heu to 0.0.5
- sf-ppu to 0.0.10.4
- FL model: up early stop from step to epoch.
- SecureAggregation uses powers of 2.
- Rename vdf partitions_dimensions to partition_columns.
- Use *args instead of args in aggregation for reducing ray task dependency.

Fixed

- FL model progress bug.
- train_test_split typo.

- Fix PPU dtype mismatch caused by JAX 32bit mode.
- Vertical PearsonR.

- FlModel/SlModel support model path dict.

- Upgrade sf-ppu version to 0.0.7.1

- More perfect HEU
- Split learning benchmark model
- SFXgboost for homo xgboost training

Added

- Support for RSA key pairs in SCQLBroker.
- Support running on [kuscia](https://github.com/secretflow/kuscia) and scheduling SCQLEngine dynamic via kuscia job.
- Added `dry_run` parameter in DoQuery request, it could be used to check query syntax and CCL without actually executing the query.
- Improve Broker high availability, support depolying in multi-node cluster deployment.
- Support reading csv from OSS/MINIO.

Changed

- **breaking**: Reshape column data type, data type `LONG` is deprecated.
- **breaking**: Modify table schema in broker storage for P2P mode.