Is the trainAsync method available for LSTM?

What is wrong?

trainAsync not available for LSTM?

Where does it happen?

• macOS v10.15.3
• VScode v1.43.2
• node v12.16.1
• brainjs v2.0.0-alpha.12

How do we replicate the issue?

(1.)

const brain = require('brain.js')
const network = new brain.recurrent.LSTM()
const jsonfileReader = require('jsonfile')

 var mappedTrainingData = await jsonfileReader
    .readFile(trainingDataPath)
    .then(res => {
      console.log('File loaded!')
      console.log('Mapping training data...')
      return res.map(elm => ({
        input: elm.title,
        output: elm.type
      }))
    })
    .then(res => {
      console.log('Mapping completed!')
      return res
    })
    .catch(error =>
      console.error(`ERROR: Error reading training data file: ${error}`)
    )
console.log('Training network...')
network
    .trainAsync(mappedTrainingData)
    .then(res => {
      console.log('Training completed!')
      networkJSON = network.toJSON()
    })

How important is this (1-5)?

2

Expected behavior (i.e. solution)

It would be great if async training is available for the other modes besides NeuralNetwork as well.

[saschamander]

@jennasisis Thx again, worked like a charm. For anyone working on this here is my code:

parent.js

exports.retrainAi = (iterations = 100, callback = undefined) => {
	const { fork } = require('child_process')

	const forked = fork('./child.js', [
		iterations.toString(),
	])

	forked.on('message', (msg) => {
		console.log(msg)
	})

	forked.on('exit', function () {
		if (callback) callback()
	})
}

child.js

const fs = require('file-system')
const brain = require('brain.js')

// Get number of iterations from args
let iterations
try {
	iterations = parseInt(process.argv[2]) || 100
} catch (error) {}

try {
	// Init network
	const network = new brain.recurrent.LSTM({})
	// Load training data
	const trainingData = require('./trainingData.json')
	// Train network
	network.train(trainingData, {
		iterations: iterations,
		log: (stats) => process.send(stats),
		logPeriod: 1,
	})
	// Save trained network
	fs.writeFileSync('./trainedNetwork.json', JSON.stringify(network.toJSON()))
} catch (error) {
	// Catch errors
	process.send(error)
} finally {
	// End child process
	process.exit()
}

[robertleeplummerjr]

Thanks @saschamander for your example. I’ve tried several times to get this working in LSTM, without success because of the complexity of the net. This I think will be a standard feature of both FeedForward and Recurrent classes when they reach stability.