text-writer-ml.js

const { Architect, Trainer } = synaptic;
const int = Math.floor;

const catagorical = i => {
    const arr = new Array(26).fill(0);
    arr[i] = 1;
    return arr;
};

Jimp.loadFont(Jimp.FONT_SANS_16_BLACK).then(function (font) {
    const trainingData = new Array(26);
    for (let i = 0; i < 26; i++) {
        const letter = String.fromCharCode(65 + i);
        const image = new Jimp(16, 16);
        image.opaque();
        image.brightness(1);
        image.print(font, 0, 0, letter);
        // image.write(`training-images/${letter}.png`);

        const d = image.bitmap.data;
        trainingData[i] = {
            input: catagorical(i),
            output: new Array(16 * 16).fill(0)
                .map((x, i) => d[i * 4] + d[i * 4 + 1] + d[i * 4 + 2])
                .map(c => c / 765)
        };
    }
    
    let neuralNetwork = new Architect.Perceptron(26, 26, 16 * 16);
    const trainer = new Trainer(neuralNetwork);
    console.log('Please wait... This might take up to 10 minutes.');
    trainer.train(trainingData, { iterations: 190 /* change to 300 for accurate results (longer loading times) */, rate: .04 }); 
    console.log('Done training!');
    
    // remove loading message
    var loadingMessage = document.getElementById('msg');
    loadingMessage.parentElement.removeChild(loadingMessage);
    
    document.addEventListener('keypress', e => {
        var code = String.fromCharCode(e.charCode).toUpperCase().charCodeAt(0);
        if (code < 65 || code > 90) return alert('Invalid letter');
        const res = neuralNetwork.activate(catagorical(code - 65));
        const col = x => Math.max(0, Math.min(255, x * 255));
        const image = new Jimp(16, 16);
        image.bitmap.data = [].concat.apply([], res.map(c => [c, c, c, 1].map(col)));
        image.getBase64(Jimp.MIME_JPEG, function (err, src) {
            var img = document.createElement("img");
            img.setAttribute("src", src);
            document.body.appendChild(img);
        });
    });
});