viadee/anchors-server repository overview

⁠javaAnchorServer

This is a server to provide Anchor-Explanations for machine learning models. It uses the Java implementation of anchors⁠ as well as adapters⁠ for accessing the data from an H2O server and optionally distributing the execution on a spark cluster. In conjunction with javaAnchorFrontend⁠ an explanation can be created and viewed easily, efficiently and fast:

A description how to read this table can be found in the ReadMe of javaAnchorFrontend⁠.

At this point, only access to data from the H2O and its models is supported. How this can be extended is described in [...].

The project is modular and each area can be easily replaced. It is also possible to extend the application so that it does not act as a server but as a command line application.

⁠Structure

The project is divided into 3 main parts:

• Access: api
• Business
• Access to data: dao
• Execution of the anchor algorithm: anchor

⁠Access

In the current version, access is via REST. This task is performed by the controller module. It accepts a request, extracts the required data from the request, and forwards the order to the business layer (business module).