DataTools4Heart Docs

Comprehensive documentation for DataTools4Heart community software

View the Project on GitHub DataTools4Heart/documentation-hub

API for Federated Processing Environment

The Federated Processing Environment (FEM) also provides an API interface that can be used to programmatically trigger distributed and federated experiments.

Authorize User

In order to use the API, authorization is required by pressing the Authorize button located at the top-right corner.

AI Dashboard - Home (Screenshot)

The required credentials are the same ones used to authenticate any user in the web interface of the Federated Processing Environment or any other DT4H product.

Basic Knowledge

In FEM, a “tool” is the core component of any experiment. It refers to a complete piece of software that can be run in a distributed or federated manner. A “tool,” therefore, can include one or more “tasks.” For example, software that is run exclusively in a distributed manner will have a single “task,” while a federated software, for instance a federated learning framework which requires at least an aggregator and one or more trainers, will have two or more “tasks.”

An “execution” is the record of a performed job. It can consist of one or more “tools” or “tasks.” For example, when running a health-check (a “tool” with a single “task”), we log an execution ID to trace the user who triggered the execution and the nodes on which it was run. Similarly, we can run a health-check tool followed by an Flcore tool (a single tool with two tasks: flcore-server and flcore-client), and all of them will be logged under a single execution ID.

Endpoints

Hosts

A “host” is a node or machine that has the FEM client installed. Therefore, it is a location where “tools” and “tasks” can be run.

Tools

A “tool” is the definition of a piece of software in FEM. A “tool” can include one or more “tasks” depending on its execution mode: distributed or federated.

Tasks

A “task” is the part of a “tool” that is executed on a “node”.

Data

Data can only be queried and downloaded by its owner.

Executions