Secretflow

- [SPU] 0.0.6.1 release
- [bugfix] Fixed several macOS build issues.

20220412

- [SPU] 0.0.6 release
- [API] Add storage_type/data_type, remove type_data.
- [Experimental] Log(natural logarithm) is calculated with Pade approximation
instead of Householder's approximation.
- [API] Fix typo, goldschmdit.
- [Improvement] fxp div, improvement accuracy with normal distributed inputs.

20220325

- [SPU] 0.0.5.1 release
- [Bugfix] Fix SEMI2K division wrong answer

20220324

- [SPU] 0.0.5 release
- [Improvement] Lift Convolution limitations.
- [Improvement] Optimize maximum/minimum speed, with MPC B2A single bit conversion.
- [Experimental] Div is calculated with direct goldschmdit method instead of
multiplication of dividend and divisor reciprocal.
- [API] change RuntimeConfig.enable_op_time_profile to enable_pphlo_profile.
- [API] change RuntimeConfig.fxp_reciprocal_goldschmdit_iters to fxp_div_goldschmdit_iters.

20220308

- [SPU] 0.0.4 release
- [Feature] add silent ot support for various ot scenarios (chosen/correlated/random
messages, chosen/correlated/random choices, 1o2/1oN)
- [Feature] add non-linear computation protocols based on silent ot (comparison,
truncation, b2a, triple, randbit, etc)
- [Feature] add a 2PC protocol: Cheetah
- [Improvement] concatenate is a lot faster
- [API] add RuntimeConfig.enable_op_time_profile to ask SPU to collect timing profiling data
- [Bugfix] fixed pphlo.ConstOp may lose signed bit after encoding

20220303

- [spu] 0.0.3 release
- [API] merge (config.proto, executable.proto, types.proto) into single spu.proto.
- [API] change RuntimeConfig.enable_protocol_trace to enable_action_trace.
- [API] change RuntimeConfig.fxp_recirptocal_goldschmdit_iters to fxp_reciprocal_goldschmdit_iters.
- [API] add RuntimeConfig.reveal_secret_condition to allow reveal secret control flow condition.
- [Bugfix] Fixed SEGV when reconstruct from an ABY3 scalar secret
- [Feature] Left/right shift now properly supports non-scalar inputs

20220217

- [spu] 0.0.2.3 release

20220211

- [spu] 0.0.2.2 release
- [Bugfix] Fix exception when kkrt psi input is too small

20220210

- [spu] 0.0.2.1 release
- [Bugfix] Fix exception when streaming psi output directory already exists

20220209

- [spu] 0.0.2 release.
- [Feature] Support multi-parties psi

20210930

- [spu] Init release.


Changelog

All notable changes to this project will be documented in this file.

**2024-05-15**

Added
- [component] support tweedie learning objective in SGB
- [component] update graph builder in model export

Added

- [component] Add stats psi component
- [component] Add score card component
- [component] Add data sampling component
- [component] Add type cast component
- [component] Optimization GLM & LR training speed by using beaver cache
- [SLModel] Add attack method BLA
- [SLModel] Add defense method CAE
- [sgb] Support sample weight training

Fix

- [component] The online prediction of the GLM model has a large deviation compared with the offline prediction.

Fix

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

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

Added

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

Removed

- No longer providing x64 macOS binary packages.