zaidalyafeai

const convlayer = tf.layers.conv2d({
  inputShape: [28, 28, 1],
  kernelSize: 5,
  filters: 8,
  strides: 1,
  activation: 'relu',
  kernelInitializer: 'VarianceScaling'
});

const input = tf.zeros([1,28,28,1]);

zaidalyafeai

const outputSize = Math.floor((inputSize-kernelSize)/stride +1);

zaidalyafeai

const LEARNING_RATE = 0.0001;
const optimizer = tf.train.adam(LEARNING_RATE);

zaidalyafeai

model.compile({
  optimizer: optimizer,
  loss: 'categoricalCrossentropy',
  metrics: ['accuracy'],
});

zaidalyafeai

const batch = tf.zeros([BATCH_SIZE,28,28,1]);
const labels = tf.zeros([BATCH_SIZE, NUM_CLASSES]);

const h = await model.fit(batch, labels,
            {
              batchSize: BATCH_SIZE,
              validationData: validationData,
              epochs: BATCH_EPOCHs
            });

zaidalyafeai

const output = tf.oneHot(tf.tensor1d([0,1,0]), 2);

//the output will be [[1, 0],[0, 1],[1, 0]]

zaidalyafeai

//h is the output of the fitting module
const loss = h.history.loss[0];
const accuracy = h.history.acc[0];

zaidalyafeai

//retrieve the canvas
const canvas = document.getElementById("myCanvas");
const ctx = canvas.getContext("2d");

//get image data
imgData = ctx.getImageData(0, 0, 28, 28);

//convert to tensor 
const tensor = tf.browser.fromPixels(imgData);

zaidalyafeai

const eTensor = tensor.expandDims(0);

zaidalyafeai

model.predict(eTensor);