TensorFlow.js Examples

This repository contains a set of examples implemented in TensorFlow.js.

Each example directory is standalone so the directory can be copied to another project.

Overview of Examples

Example name	Demo link	Input data type	Task type	Model type	Training	Inference	API type	Save-load operations
addition-rnn	🔗	Text	Sequence-to-sequence	RNN: SimpleRNN, GRU and LSTM	Browser	Browser	Layers
baseball-node		Numeric	Multiclass classification	Multilayer perceptron	Node.js	Node.js	Layers
boston-housing	🔗	Numeric	Regression	Multilayer perceptron	Browser	Browser	Layers
cart-pole	🔗		Reinforcement learning	Policy gradient	Browser	Browser	Layers	IndexedDB
custom-layer	🔗		(Defining a custom Layer subtype)			Browser	Layers
data-csv	🔗		Building a tf.data.Dataset from a remote CSV
data-generator	🔗		Building a tf.data.Dataset using a generator	Regression	Multilayer perceptron	Browser	Layers
date-conversion-attention	🔗	Text	Text-to-text conversion	Attention mechanism, RNN	Node.js	Browser and Node.js	Layers	Saving to filesystem and loading in browser
fashion-mnist-vae		Image	Generative	Variational autoencoder (VAE)	Node.js	Browser	Layers	Export trained model from tfjs-node and load it in browser
iris	🔗	Numeric	Multiclass classification	Multilayer perceptron	Browser	Browser	Layers
iris-fitDataset	🔗	Numeric	Multiclass classification	Multilayer perceptron	Browser	Browser	Layers
jena-weather	🔗	Sequence	Sequence-to-prediction	MLP and RNNs	Browser and Node	Browser	Layers
lstm-text-generation	🔗	Text	Sequence prediction	RNN: LSTM	Browser	Browser	Layers	IndexedDB
mnist	🔗	Image	Multiclass classification	Convolutional neural network	Browser	Browser	Layers
mnist-acgan	🔗	Image	Generative Adversarial Network (GAN)	Convolutional neural network; GAN	Node.js	Browser	Layers	Saving to filesystem from Node.js and loading it in the browser
mnist-core	🔗	Image	Multiclass classification	Convolutional neural network	Browser	Browser	Core (Ops)
mnist-node		Image	Multiclass classification	Convolutional neural network	Node.js	Node.js	Layers	Saving to filesystem
mnist-transfer-cnn	🔗	Image	Multiclass classification (transfer learning)	Convolutional neural network	Browser	Browser	Layers	Loading pretrained model
mobilenet	🔗	Image	Multiclass classification	Convolutional neural network		Browser	Layers	Loading pretrained model
polynomial-regression	🔗	Numeric	Regression	Shallow neural network	Browser	Browser	Layers
polynomial-regression-core	🔗	Numeric	Regression	Shallow neural network	Browser	Browser	Core (Ops)
sentiment	🔗	Text	Sequence-to-binary-prediction	LSTM, 1D convnet	Node.js or Python	Browser	Layers	Load model from Keras and tfjs-node
simple-object-detection	🔗	Image	Object detection	Convolutional neural network (transfer learning)	Node.js	Browser	Layers	Export trained model from tfjs-node and load it in browser
translation	🔗	Text	Sequence-to-sequence	LSTM encoder and decoder	Node.js or Python	Browser	Layers	Load model converted from Keras
tsne-mnist-canvas			Dimension reduction and data visualization	tSNE	Browser	Browser	Core (Ops)
webcam-transfer-learning	🔗	Image	Multiclass classification (transfer learning)	Convolutional neural network	Browser	Browser	Layers	Loading pretrained model
website-phishing	🔗	Numeric	Binary classification	Multilayer perceptron	Browser	Browser	Layers

Dependencies

Except for getting_started, all the examples require the following dependencies to be installed.

• Node.js version 8.9 or higher
• NPM cli OR Yarn

How to build an example

cd into the directory

If you are using yarn:

cd mnist-core
yarn
yarn watch

If you are using npm:

cd mnist-core
npm install
npm run watch

Details

The convention is that each example contains two scripts:

• yarn watch or npm run watch: starts a local development HTTP server which watches the filesystem for changes so you can edit the code (JS or HTML) and see changes when you refresh the page immediately.

• yarn build or npm run build: generates a dist/ folder which contains the build artifacts and can be used for deployment.

Contributing

If you want to contribute an example, please reach out to us on Github issues before sending us a pull request as we are trying to keep this set of examples small and highly curated.