convnet

ConvNetJS

ConvNetJS is a Javascript implementation of Neural networks, together with nice browser-based demos. It currently supports:

• Common Neural Network modules (fully connected layers, non-linearities)
• Classification (SVM/Softmax) and Regression (L2) cost functions
• A MagicNet class for fully automatic neural network learning (automatic hyperparameter search and cross-validatations)
• Ability to specify and train Convolutional Networks that process images
• An experimental Reinforcement Learning module, based on Deep Q Learning

For much more information, see the main page at convnetjs.com

Online Demos

• Convolutional Neural Network on MNIST digits
• Convolutional Neural Network on CIFAR-10
• Toy 2D data
• Toy 1D regression
• Training an Autoencoder on MNIST digits
• Training an Autoencoder on MNIST digits
• Deep Q Learning Reinforcement Learning demo
• Image Regression ("Painting")
• Comparison of SGD/Adagrad/Adadelta on MNIST
• MagicNet classifying arbitary CSV data demo

Example Code

Here's a minimum example of defining a 2-layer neural network and training it on a single data point:

// species a 2-layer neural network with one hidden layer of 20 neurons
var layer_defs = [];
// input layer declares size of input. here: 2-D data
// ConvNetJS works on 3-Dimensional volumes (sx, sy, depth), but if you're not dealing with images
// then the first two dimensions (sx, sy) will always be kept at size 1
layer_defs.push({type:'input', out_sx:1, out_sy:1, out_depth:2});
// declare 20 neurons, followed by ReLU (rectified linear unit non-linearity)
layer_defs.push({type:'fc', num_neurons:20, activation:'relu'}); 
// declare the linear classifier on top of the previous hidden layer
layer_defs.push({type:'softmax', num_classes:10});
 
var net = new convnetjs.Net();
net.makeLayers(layer_defs);
 
// forward a random data point through the network
var x = new convnetjs.Vol([0.3, -0.5]);
var prob = net.forward(x); 
 
// prob is a Vol. Vols have a field .w that stores the raw data, and .dw that stores gradients
console.log('probability that x is class 0: ' + prob.w[0]); // prints 0.50101
 
var trainer = new convnetjs.SGDTrainer(net, {learning_rate:0.01, l2_decay:0.001});
trainer.train(x, 0); // train the network, specifying that x is class zero
 
var prob2 = net.forward(x);
console.log('probability that x is class 0: ' + prob2.w[0]);
// now prints 0.50374, slightly higher than previous 0.50101: the networks
// weights have been adjusted by the Trainer to give a higher probability to
// the class we trained the network with (zero)

and here is a small Convolutional Neural Network if you wish to predict on images:

var layer_defs = [];
layer_defs.push({type:'input', out_sx:32, out_sy:32, out_depth:3}); // declare size of input
// output Vol is of size 32x32x3 here
layer_defs.push({type:'conv', sx:5, filters:16, stride:1, pad:2, activation:'relu'});
// the layer will perform convolution with 16 kernels, each of size 5x5.
// the input will be padded with 2 pixels on all sides to make the output Vol of the same size
// output Vol will thus be 32x32x16 at this point
layer_defs.push({type:'pool', sx:2, stride:2});
// output Vol is of size 16x16x16 here
layer_defs.push({type:'conv', sx:5, filters:20, stride:1, pad:2, activation:'relu'});
// output Vol is of size 16x16x20 here
layer_defs.push({type:'pool', sx:2, stride:2});
// output Vol is of size 8x8x20 here
layer_defs.push({type:'conv', sx:5, filters:20, stride:1, pad:2, activation:'relu'});
// output Vol is of size 8x8x20 here
layer_defs.push({type:'pool', sx:2, stride:2});
// output Vol is of size 4x4x20 here
layer_defs.push({type:'softmax', num_classes:10});
// output Vol is of size 1x1x10 here
 
net = new convnetjs.Net();
net.makeLayers(layer_defs);
 
// helpful utility for converting images into Vols is included
var x = convnetjs.img_to_vol(document.getElementById('#some_image'))
var output_probabilities_vol = net.forward(x)

Getting Started

A Getting Started tutorial is available on main page.

The full Documentation can also be found there.

See the releases page for this project to get the minified, compiled library, and a direct link to is also available below for convenience (but please host your own copy)

• convnet.js
• convnet-min.js

Compiling the library from src/ to build/

If you would like to add features to the library, you will have to change the code in src/ and then compile the library into the build/ directory. The compilation script simply concatenates files in src/ and then minifies the result.

The compilation is done using an ant task: it compiles build/convnet.js by concatenating the source files in src/ and then minifies the result into build/convnet-min.js. Make sure you have ant installed (on Ubuntu you can simply sudo apt-get install it), then cd into compile/ directory and run:

$ ant -lib yuicompressor-2.4.8.jar -f build.xml

The output files will be in build/

Use in Node

The library is also available on node.js:

1. Install it: $ npm install convnetjs
2. Use it: var convnetjs = require("convnetjs");

License

MIT