v1.6.1b0

Added

• [component] pvalue support GLM
• [component] onehot add drop parameter to support first & mode
• [component] XGB/SGB support checkpoint
• [component] LR support report weights
• [component] Add union component
• [FLModel/SLModel] tf backend support custom loss
• [SLModel] Base model support additional loss
• [SLModel] Add attack method FSHA
• [SLModel] Add defense method MID
• [SLModel] Add defense method FedPass
• [Infra] Add arm64 build support

Fixed

• [component] fix sgb export model sigmoid type not consistent with the offline prediction sigmoid type
• [component] fix sgb/xgb export model miss base score

v1.6.0b0

Added

• [component] pvalue support GLM
• [component] onehot add drop parameter to support first & mode
• [component] XGB/SGB support checkpoint
• [component] LR support report weights
• [component] Add union component
• [FLModel/SLModel] tf backend support custom loss
• [SLModel] Base model support additional loss
• [SLModel] Add attack method FSHA
• [SLModel] Add defense method MID
• [SLModel] Add defense method FedPass
• [Infra] Add arm64 build support

v1.5.0b0

Added

• SLModel/FLModel: torch backend add custom training step support.
• FLModel: Add new sample methods for datasets.
• SLModel: Add gradient_average callback.
• Component: GLM/linear-SGD Support checkpoint and recover from checkpoint.
• Component: Suppport Early Stop in SGB via XGBoost Callbacks.
• Component: Add storage interface and oss/s3 storage impl.
• Device Support cheetah h2a.

Changed

• Upgrade to Python 3.10, add support for 3.9-3.11.
• Upgrade to tensorflow 2.12.0 and pytorch 2.1.2.

Breaking Changed

• Drop support for python 3.8.

v1.5.0.dev240401

[v1.5.0.dev240401] - 2024-04-01

Added

• Add FSHA attach
• Support duplicate keys in psi comp
• Add sl/fl tf backend support custom loss

Fixed

• Resolve gpu docker install error
• Fix woe bin export err

v1.5.0.dev240319

Added

• SL: custom trainning step supported by lightning style base module.

Removed

• No longer providing x64 macOS binary packages.

v1.4.0b0

Added

• Component: Optimized memory usage and training speed for GLM.
• Component: GLM now supports early stopping base on multiple metrics.
• Component: Ability to save additional feature columns in the results of the prediction component.
• Component: Binning component can now directly display binning rules.
• Component: Provided functionality to modify binning rules.
• Component: Model export capability on secretpad, connecting offline training to online prediction.
• Component: Added split learning training and prediction component.
• SLModel: Support for GPU training in debug mode with split learning.
• SLModel: Added pipeline support and optimized DeepFM implementation with PyTorch backend.
• SLModel: Sparse compressor now supports multidimensional data compression, including images.
• SLModel: Support for inputting datasets as file names, enabling training on large datasets.
• SLModel: Introduced the AutoAttack framework for automated offense and defense, bringing multiple scenarios and attack algorithms to split learning.

Fixed

• Component: Fixed accuracy issues with GLM training.
• Infra: Fixed the issue where RayFed shutdown would hang.

v1.5.0.dev240312

Added

• SL: grad_average is supported on GPU.
• SL: mix_up defence can be called through callback.
• Simulation: add some unpartitioned datasets.

Fixed

• Fix order in componet load table.

Changed

• Serving: Update serving linkfunc type.
• FL: rename get/set_gradients methods
• SGX: improve doc for the accuracy analysis.
• SL: import fia attack.

v1.5.0.dev240304

Added

• Add gradient_average callback in SL.

Fixed

• Fix tuner default resource allocate.

Changed

• Boost the speed of autoattack benchmark.
• Upgrade to Python 3.10.

v1.4.0.dev240129

[v1.4.0.dev240129] - 2024-01-29

Added

• Split learning in PyTorch supports GPU in debug mode.

• Split learning in PyTorch supports pipeline strategy.

• Sparse compression supports the compression of multi-dimensional data, such as images.

• SLModel supports input in the form of file names and allows for multiple label inputs.

• Component: prediction operator results can now save feature columns.

• Component: binning operators can now view reports directly.

• Component: binning rule modification is supported.

• Component: support model exports, support both online and offline prediction.

Fixed

• fix: bump rayfed to 0.2.0a18 to fix shutdown lock.
• Component: improve GLM training accuracy.
• Component: reduce GLM training memory and computation cost.

v1.4.0.dev240122

Added

• Add a new feature of chunked computation to control the memory peak.
• The binning component supports displaying binning rules.

Fixed

• Fix a bug that occurs when saving the SGB model fails when the number of trees in the model is 1.
• Fix bin rule report.