Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
javascript neural networks and classifiers
branch: master

This branch is 73 commits behind harthur:master

Fetching latest commit…

Cannot retrieve the latest commit at this time

Failed to load latest commit information.
lib
test
Jakefile.js
LICENSE
README.md
package.json

README.md

brain

brain is a limited JavaScript supervised machine learning library. Full API here. Neural network example: var net = new brain.NeuralNetwork(); net.train([{input: [0, 0], output: [0]}, {input: [0, 1], output: [1]}, {input: [1, 0], output: [1]}, {input: [1, 1], output: [0]}]);

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

The output will be [0.987] or something close like that. There's no reason to use a neural network to figure out XOR, but it's a small example (-:

Naive Bayesian classifier example: var bayes = new brain.BayesianClassifier();

bayes.train("cheap replica watches", "spam");
bayes.train("I don't know if this works on Windows", "not");

var category = bayes.classify("free watches");

using as a commonJS package

To use this as a commonJS package (node/narwhal) checkout or download the code, it's a commonJS package. If you have node and npm you can:

npm install brain

then:

var brain = require("brain");
var net = new brain.NeuralNetwork();

If you didn't install with npm, you can specify the path to the brain.js file, like require("./lib/brain").

using in the browser

Download the latest brain.js. If you're using BayesianClassifier, you can only use the localStorage and (default) in-memory backends, and you'll need to grab underscore.js. If you're using the NeuralNetwork you should try to train the network offline (or on a Worker) and use the toFunction() or toJSON() options to plug the trained network in to your website.

tests

Running the tests requires node.js. To run the suite of API tests:

node test/sanity/runtests.js

cross-validation tests

The in-repo tests are just sanity/API checks, to really test out the library, run the cross-validation tests. These test the classifiers on large sets of real training data and give an error value (between 0 and 1) that indicates how good the classifier is at training. You can run the default cross-validation tests with:

node test/cvalidate/runcv.js

(requires network access to the dbs of training data). Specify your own db and options to pass in:

node test/cvalidate/runcv.js --type=neuralnetwork --db=http://localhost:5984/nndata --options='{learningRate:0.6}'

The db must be a CouchDB database of JSON objects with 'input' and 'output' fields.

Something went wrong with that request. Please try again.