Digit recognition : Training Step

testing.js

const fs = require('fs');
const db = require('node-localdb');
const csv = require('csvtojson');
const colors = require('colors');
const log = require('fancy-log');
const assert = require('assert');
const brain = require('brain.js');
const IterateObject = require("iterate-object");

const csvFilePath = 'digit_csv/train_slim.csv';

let train_data = [];
let train_data_formated = [];
let labels = [];



let contents = fs.readFileSync("training_data_output_2.json");
let jsonContent = JSON.parse(contents);
const net = new brain.NeuralNetwork()
net.fromJSON(jsonContent)


csv()
    .fromFile(csvFilePath)
    .on('done', (error) => {
        log.info('Loading training data complete !');
        generateTrainingObj();
    })
    .then((jsonObj) => {
        let jsonlength = jsonObj.length;
        for (k in jsonObj) {
            train_data.push(jsonObj[k]);
            console.log(k, jsonlength);
        }
    })

function generateTrainingObj() {

    log.info('Construct training obj..');

    for (k in train_data) {

        let label = parseInt(train_data[k].label);
        let input = [];

        IterateObject(train_data[k], function(value, name) {
            if (name != 'label') {
                input.push( parseFloat(value) / 255 );
            }
        });

        let ii = 0;
        let input_obj = input.reduce(function(o, val) { 
            o['pixel'+ii] = val; 
            ii++;
            return o; 
        }, {});

        labels.push(label);

        train_data_formated.push({ input: input_obj });

        if (parseInt(k) + 1 == train_data.length) {
            startTesting()
        }
    }


}


function startTesting() {

    log.info('Start testing !');

    for(let i = 0; i < train_data_formated.length; i++) {

        const output = net.run( train_data_formated[i].input );

        // Good result
        console.log(output, labels[i]);

    }

}

training.js

const fs = require('fs');
const db = require('node-localdb');
const csv = require('csvtojson');
const colors = require('colors');
const log = require('fancy-log');
const assert = require('assert');
const brain = require('brain.js');
const IterateObject = require("iterate-object");

/* Contain pixel color 0 - 255 */
const csvFilePath = 'digit_csv/train.csv';

let train_data = [];
let train_data_formated = [];

csv()
.fromFile(csvFilePath)
.on('done',(error)=>{
    log.info('Loading training data complete !');
    generateTrainingObj();
})
.then( (jsonObj) => {
	let jsonlength = jsonObj.length;
    for(k in jsonObj) {
    	if(k < 2000) {
	    	train_data.push( jsonObj[k] );
	    	console.log(k, jsonlength);
	    }
    }
})

function generateTrainingObj() {

	log.info('Construct training obj..');

	for(k in train_data) {

		let label = parseInt( train_data[k].label );
		let input = [];

		IterateObject(train_data[k], function (value, name) {
		    if(name != 'label') {
		    	input.push( parseFloat(value) / 255 );
		    }
		});

		let ii = 0;
		let input_obj = input.reduce(function(o, val) { 
			o['pixel'+ii] = val; 
			ii++;
			return o; 
		}, {});

		let output = null;

		if( label == 0) { 
			output = { N0: 1 }
		};

		if( label == 1) { 
			output = { N1: 1 }
		};

		if( label == 2) { 
			output = { N2: 1 }
		};

		if( label == 3) { 
			output = { N3: 1 }
		};

		if( label == 4) { 
			output = { N4: 1 }
		};

		if( label == 5) { 
			output = { N5: 1 }
		};

		if( label == 6) { 
			output = { N6: 1 }
		};

		if( label == 7) { 
			output = { N7: 1 }
		};

		if( label == 8) { 
			output = { N8: 1 }
		};

		if( label == 9) { 
			output = { N9: 1 }
		};

		train_data_formated.push( { input: input_obj, output: output } );

		console.log(parseInt(k)+1 , train_data.length)

		if(parseInt(k)+1 == train_data.length) {
			startTraining()
		}
	}

	
}


/* Note Used */
const config = {
  hiddenLayers: [300, 200, 100, 50], // array of ints for the sizes of the hidden layers in the network
  activation: 'sigmoid', // supported activation types: ['sigmoid', 'relu', 'leaky-relu', 'tanh'],
};

function startTraining() {

	log.info('Start training !');

	const net = new brain.NeuralNetworkGPU();

	net.train(train_data_formated, {
		logPeriod: 1,
		iterations: 20,
		log: detail => log.info(detail)
	});

	setTimeout(function(){

		log.info('SAVE TRAINING DATA');

		const jsonOutput = JSON.stringify( net.toJSON() );
		fs.writeFile('training_data_output_2.json', jsonOutput, 'utf8', function(){
			log.info('SAVING DONE');
		});
    
	}, 15000);

}