brain.js

About

brain.js is a library of Neural Networks written in JavaScript.

NEW! A fun and practical introduction to Brain.js

💡 Note: This is a continuation of the harthur/brain repository (which is not maintained anymore). For more details, check out this issue.

Examples

Here's an example showcasing how to approximate the XOR function using brain.js: more info on config here.

// provide optional config object (or undefined). Defaults shown.
const config = {
    binaryThresh: 0.5,
    hiddenLayers: [3],     // array of ints for the sizes of the hidden layers in the network
    activation: 'sigmoid',  // supported activation types: ['sigmoid', 'relu', 'leaky-relu', 'tanh'],
    leakyReluAlpha: 0.01   // supported for activation type 'leaky-relu'
};

// create a simple feed forward neural network with backpropagation
const net = new brain.NeuralNetwork(config);

net.train([{input: [0, 0], output: [0]},
           {input: [0, 1], output: [1]},
           {input: [1, 0], output: [1]},
           {input: [1, 1], output: [0]}]);

const output = net.run([1, 0]);  // [0.987]

or more info on config here.

// provide optional config object, defaults shown.
const config = {
    inputSize: 20,
    inputRange: 20,
    hiddenLayers: [20,20],
    outputSize: 20,
    learningRate: 0.01,
    decayRate: 0.999,
};

// create a simple recurrent neural network
const net = new brain.recurrent.RNN(config);

net.train([{input: [0, 0], output: [0]},
           {input: [0, 1], output: [1]},
           {input: [1, 0], output: [1]},
           {input: [1, 1], output: [0]}]);

const output = net.run([0, 0]);  // [0]
output = net.run([0, 1]);      // [1]
output = net.run([1, 0]);      // [1]
output = net.run([1, 1]);      // [0]

However, there is no reason to use a neural network to figure out XOR. (-: So, here is a more involved, realistic example: Demo: training a neural network to recognize color contrast.

More Examples

You can check out this fantastic screencast, which explains how to train a simple neural network using a real world dataset: How to create a neural network in the browser using Brain.js.

writing a children's book using a recurrent neural network & typescript version
using cross validation with a feed forward net & typescript version
experimental (NeuralNetwork only, but more to come!) using the gpu in a browser or using node gpu fallback to cpu & typescript version
learning math using a recurrent neural network & typescript version
predict next number, and forecast numbers & typescript version
using node streams & typescript version
simple letter detection & typescript version

Usage

Node

If you have node, you can install brain.js with npm:

npm install brain.js

Or if you prefer yarn:

yarn add brain.js

At present, the published version of brain.js is approximately 1.0.0, featuring only Feed-forward NN. All other models are beta and are being jazzed up and battle hardened. You can still download the latest, though. They are cool!

Browser

Download the latest brain.js for browser. Training is computationally expensive, so you should try to train the network offline (or on a Worker) and use the toFunction() or toJSON() options to plug the pre-trained network into your website.

Training

Use train() to train the network with an array of training data. The network has to be trained with all the data in bulk in one call to train(). More training patterns will probably take longer to train, but will usually result in a network better at classifying new patterns.

Data format

For training with `NeuralNetwork`

Each training pattern should have an input and an output, both of which can be either an array of numbers from 0 to 1 or a hash of numbers from 0 to 1. For the color contrast demo it looks something like this:

const net = new brain.NeuralNetwork();

net.train([{input: { r: 0.03, g: 0.7, b: 0.5 }, output: { black: 1 }},
           {input: { r: 0.16, g: 0.09, b: 0.2 }, output: { white: 1 }},
           {input: { r: 0.5, g: 0.5, b: 1.0 }, output: { white: 1 }}]);

const output = net.run({ r: 1, g: 0.4, b: 0 });  // { white: 0.99, black: 0.002 }

Here's another variation of the above example. (Note that input objects do not need to be similar.)

net.train([{input: { r: 0.03, g: 0.7 }, output: { black: 1 }},
           {input: { r: 0.16, b: 0.2 }, output: { white: 1 }},
           {input: { r: 0.5, g: 0.5, b: 1.0 }, output: { white: 1 }}]);

const output = net.run({ r: 1, g: 0.4, b: 0 });  // { white: 0.81, black: 0.18 }

For training with `RNNTimeStep`, `LSTMTimeStep` and `GRUTimeStep`

Each training pattern can either:

Be an array of numbers
Be an array of arrays of numbers

Example using an array of numbers:

const net = new brain.recurrent.LSTMTimeStep();

net.train([
  [1, 2, 3]
]);

const output = net.run([1, 2]);  // 3

Example using an array of arrays of numbers:

const net = new brain.recurrent.LSTMTimeStep({
  inputSize: 2,
  hiddenLayers: [10],
  outputSize: 2
});

net.train([
  [1, 3],
  [2, 2],
  [3, 1],
]);

const output = net.run([[1, 3], [2, 2]]);  // [3, 1]

For training with `RNN`, `LSTM` and `GRU`

Each training pattern can either:

Be an array of values
Be a string
Have an input and an output
- Either of which can an array of values or a string

CAUTION: When using an array of values, you can use ANY value, however, the values are represented in the neural network by a single input. So the more distinct values has the larger your input layer. If you have a hundreds, thousands, or millions of floating point values THIS IS NOT THE RIGHT CLASS FOR THE JOB. Also, when deviating from strings, this gets into beta

Example using direct strings:

const net = new brain.recurrent.LSTM();

net.train([
  'doe, a deer, a female deer',
  'ray, a drop of golden sun',
  'me, a name I call myself',
]);

const output = net.run('doe');  // ', a deer, a female deer'

Example using strings with inputs and outputs:

const net = new brain.recurrent.LSTM();

net.train([
  { input: 'I feel great about the world!', output: 'happy' },
  { input: 'The world is a terrible place!', output: 'sad' },
]);

const output = net.run('I feel great about the world!');  // 'happy'

Training Options

train() takes a hash of options as its second argument:

net.train(data, {
                            // Defaults values --> expected validation
      iterations: 20000,    // the maximum times to iterate the training data --> number greater than 0
      errorThresh: 0.005,   // the acceptable error percentage from training data --> number between 0 and 1
      log: false,           // true to use console.log, when a function is supplied it is used --> Either true or a function
      logPeriod: 10,        // iterations between logging out --> number greater than 0
      learningRate: 0.3,    // scales with delta to effect training rate --> number between 0 and 1
      momentum: 0.1,        // scales with next layer's change value --> number between 0 and 1
      callback: null,       // a periodic call back that can be triggered while training --> null or function
      callbackPeriod: 10,   // the number of iterations through the training data between callback calls --> number greater than 0
      timeout: Infinity     // the max number of milliseconds to train for --> number greater than 0
});

The network will stop training whenever one of the two criteria is met: the training error has gone below the threshold (default 0.005), or the max number of iterations (default 20000) has been reached.

By default training will not let you know how it's doing until the end, but set log to true to get periodic updates on the current training error of the network. The training error should decrease every time. The updates will be printed to console. If you set log to a function, this function will be called with the updates instead of printing to the console.

The learning rate is a parameter that influences how quickly the network trains. It's a number from 0 to 1. If the learning rate is close to 0, it will take longer to train. If the learning rate is closer to 1, it will train faster, but training results may be constrained to a local minimum and perform badly on new data.(Overfitting) The default learning rate is 0.3.

The momentum is similar to learning rate, expecting a value from 0 to 1 as well, but it is multiplied against the next level's change value. The default value is 0.1

Any of these training options can be passed into the constructor or passed into the updateTrainingOptions(opts) method and they will be saved on the network and used during the training time. If you save your network to json, these training options are saved and restored as well (except for callback and log, callback will be forgotten and log will be restored using console.log).

A boolean property called invalidTrainOptsShouldThrow is set to true by default. While the option is true, if you enter a training option that is outside the normal range, an error will be thrown with a message about the abnormal option. When the option is set to false, no error will be sent, but a message will still be sent to console.warn with the related information.

Async Training

trainAsync() takes the same arguments as train (data and options). Instead of returning the results object from training, it returns a promise that when resolved will return the training results object.

  const net = new brain.NeuralNetwork();
  net
    .trainAsync(data, options)
    .then(res => {
      // do something with my trained network
    })
    .catch(handleError);

With multiple networks you can train in parallel like this:

  const net = new brain.NeuralNetwork();
  const net2 = new brain.NeuralNetwork();

  const p1 = net.trainAsync(data, options);
  const p2 = net2.trainAsync(data, options);

  Promise
    .all([p1, p2])
    .then(values => {
      const res = values[0];
      const res2 = values[1];
      console.log(`net trained in ${res.iterations} and net2 trained in ${res2.iterations}`);
      // do something super cool with my 2 trained networks
    })
    .catch(handleError);

Cross Validation

Cross Validation can provide a less fragile way of training on larger data sets. The brain.js api provides Cross Validation in this example:

const crossValidate = new brain.CrossValidate(brain.NeuralNetwork, networkOptions);
crossValidate.train(data, trainingOptions, k); //note k (or KFolds) is optional
const json = crossValidate.toJSON(); // all stats in json as well as neural networks
const net = crossValidate.toNeuralNetwork(); // get top performing net out of `crossValidate`


// optionally later
const json = crossValidate.toJSON();
const net = crossValidate.fromJSON(json);

Use CrossValidate with these classes:

brain.NeuralNetwork
brain.RNNTimeStep
brain.LSTMTimeStep
brain.GRUTimeStep

An example of using cross validate can be found in examples/cross-validate.js

Train Stream

Streams are a very powerful tool in node for massive data spread across processes and are provided via the brain.js api in the following way:

const net = new brain.NeuralNetwork();
const trainStream = new brain.TrainStream({
  neuralNetwork: net,
  floodCallback: function() {
    flood(trainStream, data);
  },
  doneTrainingCallback: function(stats) {
    // network is done training!  What next?
  }
});

// kick it off
readInputs(trainStream, data);

function readInputs(stream, data) {
  for (let i = 0; i < data.length; i++) {
    stream.write(data[i]);
  }
  // let it know we've reached the end of the inputs
  stream.endInputs();
}

An example of using train stream can be found in examples/stream-example.js

Methods

`train(trainingData)` -> trainingStatus

The output of train() is a hash of information about how the training went:

{
  error: 0.0039139985510105032,  // training error
  iterations: 406                // training iterations
}

`run(input)` -> prediction

Supported on classes:

brain.NeuralNetwork
brain.NeuralNetworkGPU -> All the functionality of brain.NeuralNetwork but, ran on GPU (via gpu.js in WebGL2, WebGL1, or fallback to CPU)
brain.recurrent.RNN
brain.recurrent.LSTM
brain.recurrent.GRU
brain.recurrent.RNNTimeStep
brain.recurrent.LSTMTimeStep
brain.recurrent.GRUTimeStep

Example:

// feed forward
const net = new brain.NeuralNetwork();
net.fromJSON(json);
net.run(input);

// time step
const net = new brain.LSTMTimeStep();
net.fromJSON(json);
net.run(input);

// recurrent
const net = new brain.LSTMTimeStep();
net.fromJSON(json);
net.run(input);

`forecast(input, count)` -> predictions

Available with the following classes. Outputs a array of predictions. Predictions being a continuation of the inputs.

brain.recurrent.RNNTimeStep
brain.recurrent.LSTMTimeStep
brain.recurrent.GRUTimeStep

Example:

const net = new brain.LSTMTimeStep();
net.fromJSON(json);
net.forecast(input, 3);

`toJSON() -> json`

Serialize neural network to json

`fromJSON(json)`

Deserialize neural network from json

Failing

If the network failed to train, the error will be above the error threshold. This could happen if the training data is too noisy (most likely), the network does not have enough hidden layers or nodes to handle the complexity of the data, or it has not been trained for enough iterations.

If the training error is still something huge like 0.4 after 20000 iterations, it's a good sign that the network can't make sense of the given data.

RNN, LSTM, or GRU Output too short or too long

The instance of the net's property maxPredictionLength (default 100) can be set to adjust the output of the net;

Example:

const net = new brain.recurrent.LSTM();

// later in code, after training on a few novels, write me a new one!
net.maxPredictionLength = 1000000000; // Be careful!
net.run('Once upon a time');

JSON

Serialize or load in the state of a trained network with JSON:

const json = net.toJSON();
net.fromJSON(json);

Standalone Function

You can also get a custom standalone function from a trained network that acts just like run():

const run = net.toFunction();
const output = run({ r: 1, g: 0.4, b: 0 });
console.log(run.toString()); // copy and paste! no need to import brain.js

Options

NeuralNetwork() takes a hash of options:

const net = new brain.NeuralNetwork({
  activation: 'sigmoid', // activation function
  hiddenLayers: [4],
  learningRate: 0.6 // global learning rate, useful when training using streams
});

activation

This parameter lets you specify which activation function your neural network should use. There are currently four supported activation functions, sigmoid being the default:

sigmoid
relu
leaky-relu
- related option - 'leakyReluAlpha' optional number, defaults to 0.01
tanh

Here's a table (thanks, Wikipedia!) summarizing a plethora of activation functions — Activation Function

hiddenLayers

You can use this to specify the number of hidden layers in the network and the size of each layer. For example, if you want two hidden layers - the first with 3 nodes and the second with 4 nodes, you'd give:

hiddenLayers: [3, 4]

By default brain.js uses one hidden layer with size proportionate to the size of the input array.

Streams

The network now has a WriteStream. You can train the network by using pipe() to send the training data to the network.

Example

Refer to stream-example.js for an example on how to train the network with a stream.

Initialization

To train the network using a stream you must first create the stream by calling net.createTrainStream() which takes the following options:

floodCallback() - the callback function to re-populate the stream. This gets called on every training iteration.
doneTrainingCallback(info) - the callback function to execute when the network is done training. The info param will contain a hash of information about how the training went:

{
  error: 0.0039139985510105032,  // training error
  iterations: 406                // training iterations
}

Transform

Use a Transform to coerce the data into the correct format. You might also use a Transform stream to normalize your data on the fly.

Utilities

`likely`

const likely = require('brain/likely');
const key = likely(input, net);

Likely example see: simple letter detection

`toSVG`

<script src="../../src/utilities/svg.min.js"></script>

Renders the network topology of a feedforward network

document.getElementById('result').innerHTML = brain.utilities.toSVG(network,options)

toSVG example see: network rendering

The user interface used:

Neural Network Types

brain.NeuralNetwork - Feedforward Neural Network with backpropagation
brain.NeuralNetworkGPU - Feedforward Neural Network with backpropagation, GPU version
brain.recurrent.RNNTimeStep - Time Step Recurrent Neural Network or "RNN"
brain.recurrent.LSTMTimeStep - Time Step Long Short Term Memory Neural Network or "LSTM"
brain.recurrent.GRUTimeStep - Time Step Gated Recurrent Unit or "GRU"
brain.recurrent.RNN - Recurrent Neural Network or "RNN"
brain.recurrent.LSTM - Long Short Term Memory Neural Network or "LSTM"
brain.recurrent.GRU - Gated Recurrent Unit or "GRU"

Why different Neural Network Types?

Different neural nets do different things well. For example:

A Feedforward Neural Network can classify simple things very well, but it has no memory of previous actions and has infinite variation of results.
A Time Step Recurrent Neural Network remembers, and can predict future values.
A Recurrent Neural Network remembers, and has a finite set of results.

Get Involved!

Issues

If you have an issue, either a bug or a feature you think would benefit your project let us know and we will do our best.

Create issues here and follow the template.

Contributors

This project exists thanks to all the people who contribute. [Contribute].

Backers

Thank you to all our backers! 🙏 [Become a backer]

Name		Name	Last commit message	Last commit date
Latest commit History 665 Commits
dist		dist
examples-typescript		examples-typescript
examples		examples
src		src
test		test
.babelrc		.babelrc
.gitignore		.gitignore
.npmignore		.npmignore
.travis.yml		.travis.yml
CONTRIBUTING.md		CONTRIBUTING.md
ISSUE_TEMPLATE.md		ISSUE_TEMPLATE.md
LICENSE		LICENSE
PULL_REQUEST_TEMPLATE.md		PULL_REQUEST_TEMPLATE.md
README.md		README.md
bower.json		bower.json
browser.js		browser.js
browser.min.js		browser.min.js
index.d.ts		index.d.ts
index.js		index.js
logo.svg		logo.svg
package-lock.json		package-lock.json
package.json		package.json

License

sygops/brain.js

Folders and files

Latest commit

History

Repository files navigation

brain.js

About

Table of Contents

Examples

More Examples

Usage

Node

Browser

Training

Data format

For training with NeuralNetwork

For training with RNNTimeStep, LSTMTimeStep and GRUTimeStep

For training with RNN, LSTM and GRU

Training Options

Async Training

Cross Validation

Train Stream

Methods

train(trainingData) -> trainingStatus

run(input) -> prediction

forecast(input, count) -> predictions

toJSON() -> json

fromJSON(json)

Failing

RNN, LSTM, or GRU Output too short or too long

JSON

Standalone Function

Options

activation

hiddenLayers

Streams

Example

Initialization

Transform

Utilities

likely

toSVG

Neural Network Types

Why different Neural Network Types?

Get Involved!

Issues

Contributors

Backers

Sponsors

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Languages

For training with `NeuralNetwork`

For training with `RNNTimeStep`, `LSTMTimeStep` and `GRUTimeStep`

For training with `RNN`, `LSTM` and `GRU`

`train(trainingData)` -> trainingStatus

`run(input)` -> prediction

`forecast(input, count)` -> predictions

`toJSON() -> json`

`fromJSON(json)`

`likely`

`toSVG`

Packages