We will release FATE 2.0.0-alpha version in Feb

[dylan-fan]

Branch

• FATE： https://github.com/FederatedAI/FATE/tree/release-2.0-alpha
• FATE-Flow：https://github.com/FederatedAI/FATE-Flow/tree/release-2.0-alpha
• FATE-Client：https://github.com/FederatedAI/FATE-Client/tree/release-2.0-alpha
• Eggroll：https://github.com/WeBankFinTech/eggroll/tree/release-2.5.0-alpha

Release 2.0.0-alpha

Feature Highlights

Arch 2.0：Building Unified and Standardized API for Heterogeneous Computing Engines Interconnection

• Introduce Context to manage useful APIs for developers, such as Metrics, Cipher, Tensor and IO.
• Introduce Tensor data structure to handle local and distributed matrix operation, with built-in heterogeneous acceleration support.
• Introduce DataFrame, a 2D tabular data structure for data io and simple feature engineering.
• Refactor logger, customizable logging for different use cases and flavors.
• Introduce new high-level federation API suite: context.<role>.get(name)/context.<role>.put(name=value).

Components 2.0: Building Standardized Algorithm Components for different Scheduling Engines

• Introduce components toolbox to wrap ML modules as standard executable programs.
• spec and loader expose clear API for smooth internal extension and external system integration.
• Provide several cli tools to interact and execute components.
• Implement base demos components: reader, intersection, feature scale, lr and evaluation.

ML 2.0(demo)

• Provide base demos for federated machine learning algorithm: intersection、feature scale、lr and evaluation.

Pipeline 2.0: Building Scalable Federated DSL for Application Layer Interconnection

• Introduce new scalable and standardized federated DSL IR(Intermediate Representation) for federated modeling job
• Compile python client to DSL IR
• Support multiple scalable execution backends, including standalone and Fate-Flow.

OSX(Open Site Exchange) 1.0: Building Open Platform for Cross-Site Communication Interconnection

• Standardized Cross-Site lower-level federation api
• Support grpc synchronous transmission and streaming transmission; Compatible with eggroll interface and can replace FATE-1.x rollsite component
• Support asynchronous message transmission, which can replace rabbitmq and pulsar components in FATE-1.x
• Support HTTP-1.X protocol transmission
• Support cluster deployment and inter-site traffic control
• Support networking as an Exchange component

FATE Flow 2.0: Building Open and Standardized Scheduling Platform for Scheduling Interconnection

• Adapted to new scalable and standardized federated DSL IR
• Standardized API interface with param type checking
• Decoupling Flow from FATE repository
• Optimized scheduling logic, with configurable dispatcher decoupled from initiator
• Support container-level algorithm loading and task scheduling, enhancing support for cross-platform heterogeneous scenarios
• Independent maintenance for system configuration to enhance flexibility and ease of configuration
• Support new communication engine OSX, while compatible with all engines from Flow 1.X
• Introduce OFX(Open Flow Exchange) module: encapsulated scheduling client to allow cross-platform scheduling

Deploy

• Support installing from PyPI

[dylan-fan]

LGTM

[LaynePeng]

LGTM

[excellent-lixl]

LGTM

[hainingzhang]

LGTM

[snowzjx]

LGTM