BRAIN TEXT

This is a wrapper module over brain.js intended to perform text classification.

The module uses Bag Of Word strategy and One Shot Encoding to map text and labels (classes) to vectors suitable to be fed in an artificial neural network of brain.js library.

How to use

First create an BrainText object:

let brainText = new BrainText();

The network will be trained with this default configuration:

{
    iterations: 3000, // the maximum times to iterate the training data
    errorThresh: 0.006, // the acceptable error percentage from training data
    log: true, // true to use console.log, when a function is supplied it is used
    logPeriod: 10, // iterations between logging out
    learningRate: 0.3, // multiply's against the input and the delta then adds to momentum
    momentum: 0.1, // multiply's against the specified "change" then adds to learning rate for change
};

but it can be changed if you want by means of setConfiguration:

brainText.setConfiguration(config);

The input must be a JSON string with labels and texts. The format must be like this:

const modelJSON = '{"encender_lampara": ["enciende la luz","esto está muy oscuro"],"apagar_lampara": ["apaga la luz","apaga la lámpara"]}';

This data can be loaded in network by means of:

brainText.loadTrainDataFromInputDataString(modelJSON);

And now the network can be trained:

let result = brainText.train();

Since training can be a very intensive CPU operation and may last for a while, the function train() returns a promise. You only should use the network to perform classification once the trainig process is finished. You can do that by using run() function, like this:

result.then(() => {
    let r = brainText.run("encender luz");
    console.log(r);
});

The result has this aspect:

{ text: 'encender luz',
  label: 'encender_lampara',
  confidence: 0.7557198405265808,
  prediction: 
   { encender_lampara: 0.7557198405265808,
     apagar_lampara: 0.24401657283306122 },
     status: 'TRAINED' 
   }

After traininig, new data can be added to train data:

brainText.addData([
    {label: 'encender_lampara', text: 'dale a la lamparita'},
    {label: 'apagar_lampara', text: 'quita la lamparita'}
    ]);

If you want to add just a train data:

brainText.addOneData({label: 'encender_lampara', text: 'dale a la lamparita'});

Now the network is OUTDATED. To take into account these new data, the network must be trained again:

let result = brainText.train();

We can obtain a JSON model

let jsonModel = brainText.toJSON();

This json model can be used to be loaded as model in a BrainText object:

let jsonModel = brainText.toJSON();

let brainText2 = new BrainText();

brainText2.fromJSON(jsonModel);

Once a new BrainText object has been created by means of fromJSON, you can add new data and train again to obtain a better model.

This JSON model can be also used in order to create a neural network with brain.js library:

let jsonModel = brainText.train();

const brain = require('brain.js');

let net = new new brain.NeuralNetwork();

net.fromJSON(jsonModel.net);

The file test/test.jsshows how to use this module.