@epfml/discojs v2.1.1

Disco.js Web Module

discojs-web contains the browser-only code of Disco.js, based off and extending discojs-core.

Installation

The discojs-web project is available as the @epfml/discojs NPM package, which can be installed with npm i @epfml/discojs.

Development Environment

The dev tools run on Node.js and require npm, a package manager for the Node.js runtime environment. We recommend using nvm for installing both Node.js and NPM.

To install the project's dependencies, run:

cd ..
npm ci

Since the dependencies of discojs-core, discojs-web and discojs-node are the same, they are specified in a top-level package.json file, to ease installation and building.

⚠ WARNING: Apple Silicon. TensorFlow.js version 3 do support M1 processors for macs. To do so, make sure you have an arm Node.js executable installed (not x86_64). It can be checked using:

node -p "process.arch"

which should return something similar to arm64.

Build

The server and CLI modules, as well as all unit tests (except Cypress) use the discojs-web interface, i.e. they all run on Node.js. This Disco.js Node module is build on top of and extends discojs-core, whose code is symlinked into discojs-web/src/core. To build this project:

npm run build

This invokes the TypeScript compiler (tsc). It will output the compilation files of discojs-web in a dist/ directory. To recompile from stratch, simply rm -rf dist/ before running npm run build again.

Development

Contributing

Contributions to discojs-web must only include browser-specific code. Code common to both the browser and Node.js must be added to discojs-core instead.

As a rule of thumb, the src/core/ directory must never be modified when modifying discojs-web, since it is symlinked to discojs-core.

If you wish to add a new file or submodule to the project, please do so in a similar way as src/core/ is structured. That is, adding a new task to discojs-web would mean adding a new file to src/tasks/ and modifying src/tasks/index.ts (NOT src/core/tasks/...).

Note that, if you end up making calls to the Tensorflow.js API, you must import it from the root index. This is to ensure the Node version of TF.js is loaded, and only once.