TensorFlow.js is a Javascript library to train and deploy ML models on browsers and Node.js. It takes several steps to set it up on Ubuntu 20.04

Step 1. Install node.js and npm

sudo apt-get install nodejs
sudo apt-get install npm

Step 2. Initiate a node.js project

npm init

Step 3. Install TensorFlow.js

npm install @tensorflow/tfjs
npm install @tensorflow/tfjs-node # install with C++ bindings

Step 4. Test with examples

I tested the installation with an official case.

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

// 可选加载绑定：
// 如果使用GPU运行，请使用'@tensorflow/tfjs-node-gpu'
require('@tensorflow/tfjs-node');

// 训练一个简单模型:
const model = tf.sequential();
model.add(tf.layers.dense({units: 100, activation: 'relu', inputShape: [10]}));
model.add(tf.layers.dense({units: 1, activation: 'linear'}));
model.compile({optimizer: 'sgd', loss: 'meanSquaredError'});

const xs = tf.randomNormal([100, 10]);
const ys = tf.randomNormal([100, 1]);

model.fit(xs, ys, {
  epochs: 100,
  callbacks: {
    onEpochEnd: (epoch, log) => console.log(`Epoch ${epoch}: loss = ${log.loss}`)
  }
});

Save a file called test.js and run with node test.js.

verriding the gradient for 'Max'
Overriding the gradient for 'OneHot'
Overriding the gradient for 'PadV2'
Overriding the gradient for 'SpaceToBatchND'
Overriding the gradient for 'SplitV'
node-pre-gyp info This Node instance does not support builds for N-API version 6
node-pre-gyp info This Node instance does not support builds for N-API version 6
2020-07-15 17:54:37.755169: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-07-15 17:54:37.793073: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2195835000 Hz
2020-07-15 17:54:37.793702: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x1246000 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-07-15 17:54:37.793726: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
Epoch 1 / 100
eta=0.0 ==============================================================================================> 
81ms 806us/step - loss=1.25 
Epoch 0: loss = 1.254112720489502
Epoch 2 / 100
eta=0.0 ==============================================================================================> 
34ms 341us/step - loss=1.23 
Epoch 1: loss = 1.2288624048233032
Epoch 3 / 100

If there are errors like

/home/dsk/xxx/tensorboard/fuzzer_dsk/node_modules/@tensorflow/tfjs-node/dist/index.js:49
    throw new Error("The Node.js native addon module (tfjs_binding.node) can not "
    ^

Error: The Node.js native addon module (tfjs_binding.node) can not be found at path: /home/dsk/xxx/tensorboard/fuzzer_dsk/node_modules/@tensorflow/tfjs-node/lib/napi-v5/tfjs_binding.node. 
Please run command 'npm rebuild @tensorflow/tfjs-node build-addon-from-source' to rebuild the native addon module.

Run npm rebuild @tensorflow/tfjs-node build-addon-from-source as told.

If you met python errors during this saying gyp==0.1 distribution was not found, run pip install git+https://chromium.googlesource.com/external/gyp as shown in this.

Test another case from official document

import * as tf from '@tensorflow/tfjs';
import {loadGraphModel} from '@tensorflow/tfjs-converter';

const MODEL_URL = 'model_directory/model.json';

const model = await loadGraphModel(MODEL_URL);
const cat = document.getElementById('cat');
model.execute(tf.fromPixels(cat));

Save it as test.ts and compile it with tsc test.ts.

If there are errors saying Accessors are only available when targeting ECMAScript 5 and higher, use tsc -t es5 test.ts instead.

If there are errors around “WebGL2RenderingContext compatibility”, like this article shows, run npm -i --save @types/webgl2 to fix it.

If there are errors like Build: Cannot find type definition file for 'node', run npm install @types/node --save-dev as shown in this.

After solving all issues, you will get a test.js file, then you can run it with node test.js.