TensorFlow.js

JavaScript™ belongs to Oracle

Deep Learning is all about linalg

JavaScript sucks at linalg

JavaScript no more sucks at linalg

JavaScript even has autodiff now

Quick facts

• Started in 2017 as deeplearn.js
• Written in TypeScript
• ~10 developers from Google (Brain)

Official features

• Develop ML in the browser
• Run existing models
• Retrain existing models

Run Existing models

API levels 

{

CoreKeras

Core API

Core API

• Copied from Python; naming convention changed
• Eager-only
  • No device pinning/selection
• A few quirks
  • There are API differences
  • No tf.nn
  • Semi-manual memory management
  • "web friendly" model format

Tensorflow model formats zoo

1. GraphDef = graph + [variable values - 2GB limit]
2. Checkpoint = variable values
3. Summary (aka Tensorboard) = GraphDef + key-value
4. MetaGraph = GraphDef + tags + i/o + Checkpoint
5. SavedModel = one or more MetaGraph-s
6. ModuleDef (aka Hub) = special SavedModel
7. Keras = hdf5 with arch and weights
8. 🎉 TensorFlow.js = GraphDef + JSON + sharded weights

Semi-manual memory management

• There are various JS engines
• There are various GC implementations
• We allocate memory like hungry 🐺

Semi-manual memory management

const y = tf.tidy(() => {
   const one = tf.scalar(1);
   const a = tf.scalar(2);
   const b = a.square();
   console.log('numTensors (in tidy): ' + tf.memory().numTensors);
   return b.add(one);
});
console.log('numTensors (outside tidy): ' + tf.memory().numTensors);
y.print();  // y = 2 ^ 2 + 1
        

Tensor IO

• tf.Tensor is immutable
• tf.TensorBuffer is mutable
• No efficient set()

Keras API

Keras is not tf.keras

Keras API

• No CuDNN
• Tensors may leak
• Advanced parameters not supported

Mobilenet

Progressive GAN

Transfer Learning

Examples

• Human pose recognition
• Playing pacman with gestures
• Translation (seq2seq)
• Text classification
• Many more!

Installation for Node.js

npm install @tensorflow/tfjs @tensorflow/tfjs-node

                          or @tensorflow/tfjs-node-gpu

ijavascript

npm install ijavascript-await

• Node.js driver for Jupyter
• Deals with async/await
• Good integration into the notebook

Summary

• Much fun and easy
• Very early days
• PoCs work
• TensorFlow is getting even more complex inside