@fedml/spider

A Lightweight and Customizable Federated Learning Javascript SDK for Web Browser, Backed by FedML MLOps

Usage

• Install it with any node module manager

  # npm
  npm install @fedml/spider -S
  # yarn
  yarn add @fedml/spider -S
  # pnpm
  pnpm install @fedml/spider -S

• Install peer dependencies

  npm i @tensorflow/tfjs -S

• Import fedml_train function to start train

  import { fedml_train } from '@fedml/spider'
  
  // prepare running args 
  const client_id = YOUR_CLIENT_ID
  const run_args = await AnyFunctionFetchRunArgs(...)
  
  // start training
  fedml_train(run_args, client_id, {
    // customDataLoader?: <Optional: Your custom data loader>
  })

• API

  type DataLoader = (
    args: any,
    client_id: any,
    trainBatchSize: number,
    testBatchSize: number
  ) => Promise<{
    trainData: any
    trainDataLabel: any
    testData: any
    testDataLabel: any
  }>
  
  interface Options {
    customDataLoader?: DataLoader
  }
  
  function fedml_train(
    run_args: any,
    client_id: string | number,
    options?: Options
  ): Promise<void>

Contributing

We recommend that you use pnpm(https://pnpm.io/installation) as the package manager.

catalog

• src: Here the source codes, and src/index.ts is the entry file for rollup bundle.
• dist: The rollup bundle artifacts is output here. This will be not be committed to the GitHub codebase, however, will be released to npm.

script

• pnpm install: Bootstrap dependencies of this project.
• pnpm run dev: Starting rollup bundler with watch mode.
• pnpm run lint:fix: Format source code style under src folder with eslint and prettier.
• pnpm run release(Authorization required ): Release a upgraded version package to npm, then mark and push a released tag to GitHub.