Available JS Libraries for ML

1. Tensorflow.js

Setting up Tensorflow.js

1. Create a directory and open terminal in that directory
2. Run

npm init

3. Choose defaults
4. Now Install tensorflow.

npm i @tensorflow/tfjs

5. In order to execute code in browser you need a bundler as well.(Parcel is recommended for that purpose). Run

npm install -g parcel-bundler --save-dev

This will include parcel as a dev dependency

6. Now Add the following in your scripts section in package.json

"dev": "parcel <your entry file>",
"build": "parcel build <your entry file>"

You can check if your bowser supports WebGL by going on WebGL and checking if the cube spins.

By default, if your browser supports WEBGL, then you can check the backend using following code.

console.log(tf.version)

tf.ready().then(() => {

    console.log(tf.getBackend());

});

What is WebGL?

WebGL is a JavaScript API for rendering interactive 2D and 3D graphics within any compatible web browser without the use of plug-ins. WebGL is fully integrated with other web standards, allowing GPU-accelerated usage of physics and image processing and effects as part of the web page canvas.

WASM Backend

In order to set wasm backend i.e WebAssembly, then you need to follow following steps

1. Run following command in cmd

npm i @tensorflow/tfjs-backend-wasm

2. Import it in index.js file

import '@tensorflow/tfjs-backend-wasm';

3. Now set the backend as wasm.

tf.setBackend('wasm');

4. Now, you need a parcel plugin to copy the staic files inside its node_modules directory to the dist folder of application

npm i -D parcel-plugin-static-files-copy

This will include static-files-copy as your dev dependency

5. Now you need to add the following code in package.json file

"staticFiles": {
    "staticPath": "./node_modules/@tensorflow/tfjs-backend-wasm/dist",
    "excludeGlob": ["**/!(*.wasm)"]
  },

WASM is good option if you want too target low end devices with no GPU and less processing power.

Server Side Tensorflow

If your use case demands running your models on server side, then you can use nodeJS for that.

Run the following command on cmd

npm i @tensorflow/tfjs-node

Now that we have server side bindings for tensorflow, we can test it out.

1. Create a file server.js
2. Paste following code🥇

const tf = require('@tensorflow/tfjs');
require('@tensorflow/tfjs-node');

console.log(tf.version);
tf.ready().then(() => {
  console.log(tf.getBackend());
});

In windows, you will run into errors as it will ask for for dependencies, However in Mac Big Sur it worked perfectly fine.

NodeJS can also be used to make API endpoints as well. The most popular framework for this use case is ExpressJS.

If you want to expose your tensorflowJS code as API, either for training or for making predictions, you can use ExpressJS.

npm i express

Now you can make an app using following code🥇

const express = require('express');
const app = express();

Create an endpoint and call it "train"

app.get('/train', function(req, res){

// demo code < you can include training code here>

console.log(tf.version);
tf.ready().then(()=>{
  const msg = `Loaded Tensorflow.js-version is ${tf.version} with backend ${tf.getBackend()}`;
  console.log(msg);
  res.send(msg);
})

app.listen(9000,function(req,res){
  console.log('Running server on port 9000 ...');
})

});

Now go to url localhost:9000/train

And you will see the response from the api on the browser webpage.