Titanic: Machine Learning from Disaster with brain.js

Titanic: Machine Learning from Disaster with brain.js

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

const trainPath = './train.csv';
const testPath = './test.csv';
const gp = './gpinr.csv';

var train = ezcsv.csv.read(trainPath);
var test = ezcsv.csv.read(testPath);
var gpinr = ezcsv.csv.read(gp);

var headerTrain = ['PassengerId', 'Survived', 'Pclass', 'Name', 'Sex', 'Age', 'SibSp', 'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'];

var headerTest = ['PassengerId', 'Pclass', 'Name', 'Sex', 'Age', 'SibSp', 'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'];




function encode(arg) {
    if (typeof arg == "number")
        return arg;
    else
        return arg.split('').map(x => (x.charCodeAt(0) / 255));
}

function normalization(val) {
    var lres;

    if (val > 1 && val < 9) {
        lres = (val / 10);
    }
    if (val > 9 && val < 99) {
        lres = (val / 100);
    }
    if (val > 99 && val < 999) {
        lres = (val / 1000);
    }
    if (val > 999 && val < 9999) {
        lres = (val / 10000);
    }
    if (val > 9999 && val < 99999) {
        lres = (val / 100000);
    }
    if (val > 99999 && val < 999999) {
        lres = (val / 1000000);
    }
    if (val > 999999 && val < 9999999) {
        lres = (val / 10000000);
    }

    if (lres == null || lres === undefined) {
        return 0;
    }
    return lres;

}

function checkNan(val) {
    return isNaN(val) ? 0 : normalization(val);
}

function strEmpty(val) {
    return (val == '') ? 0 : encode(val)[0];
}
const brainForm = (input) => {
    let temp = {
        input: {},
        output: {}
    };
    //console.log(input);
    // temp.input[headerTrain[0]] = parseInt(input.PassengerId);
    temp.input[headerTrain[2]] = checkNan(parseInt(input.Pclass));
    //temp.input[headerTrain[3]] = encode(input.Name);
    //temp.input[headerTrain[4]] = encode(input.Sex);
    // temp.input[headerTrain[5]] = parseInt(input.Age);
    temp.input[headerTrain[6]] = checkNan(parseInt(input.SibSp));
    temp.input[headerTrain[7]] = checkNan(parseInt(input.Parch));
    temp.input[headerTrain[8]] = checkNan(parseInt(input.Ticket));
    temp.input[headerTrain[9]] = checkNan(parseFloat(input.Fare));
    temp.input[headerTrain[10]] = normalization(parseFloat(input.Cabin));
    temp.input[headerTrain[11]] = strEmpty(input.Embarked);
    temp.output['Survived'] = parseInt(input.Survived);
    return temp;

}

const testForm = input => {
    let temp = {

    };

    //temp[headerTrain[0]] = parseInt(input.PassengerId);
    temp[headerTest[1]] = checkNan(parseInt(input.Pclass));
    //temp[headerTest[2]] = encode(input.Name);
    //temp[headerTest[3]] = encode(input.Sex);
    //temp[headerTest[4]] = parseInt(input.Age);
    temp[headerTest[5]] = checkNan(parseInt(input.SibSp));
    temp[headerTest[6]] = checkNan(parseInt(input.Parch));
    temp[headerTest[7]] = checkNan(parseInt(input.Ticket));
    temp[headerTest[8]] = checkNan(parseFloat(input.Fare));
    temp[headerTest[9]] = normalization(parseFloat(input.Cabin));
    temp[headerTest[10]] = strEmpty(input.Embarked);


    return temp;
}


let brainData = [];

train.items.map((res, key) => {
    brainData.push(brainForm(res));
});


var config = {
    iterations: 5000,
    log: true,
    logPeriod: 500,
    errorThresh: 0.003,
    hiddenLayers: [3, 3],

    learningRate: 0.3,
    activation: 'sigmoid'
}
var net = new brain.NeuralNetwork(config);


net.train(brainData);

let lastPassengerId = parseInt(train.items[train.items.length - 1].PassengerId) + 1;

let results = [];

test.items.map((res, key) => {

    let temp = {
        PassengerId: lastPassengerId + key,
        Survived: Math.round(net.run(testForm(res)).Survived)
    };

    results.push(temp);

});

var matches = 0;
for (let i = 0; i < results.length; i++) {
    if (results[i].PassengerId == parseInt(gpinr.items[i].PassengerId) && results[i].Survived == parseInt(gpinr.items[i].Survived))
        matches++;
}

console.log(`Score: ${matches}/${results.length}`);


var submission = ezcsv.csv.write(
    'titanicSubmission.csv', {
        header: ['PassengerId', 'Survived'],
        items: results
    }
);