Marina-Miranovich

• TDD: https://en.wikipedia.org/wiki/Test-driven_development
• TDD profit graphs http://www.webcitation.org/61G4kow2C and http://www.agilemodeling.com/essays/costOfChange.htm
• Unit testing: https://en.wikipedia.org/wiki/Unit_testing
• Test classification: http://habrahabr.ru/company/npo-comp/blog/223833/
• Testing mind map: https://github.com/polarnik/TypesOfTesting

And a ton of literature that doesn't directly mentioned in the presentation, but it still has a big influence on my speech:

• http://habrahabr.ru/post/119465/ (rus)
• https://blog.8thlight.com/uncle-bob/2014/04/30/When-tdd-does-not-work.html (eng)
• http://img105.job1001.com/upload/adminnew/2015-02-05/1423102064-STSV0KT.pdf (eng and book!)

Marina-Miranovich

np.random.seed(44)
ia.seed(44)
def main():
    for i in range(1, 191):
        draw_single_sequential_images(str(i), "others", "others-aug")
    for i in range(1, 191):
        draw_single_sequential_images(str(i), "hits", "hits-aug")
    for i in range(1, 191):
        draw_single_sequential_images(str(i), "kicks", "kicks-aug")
def draw_single_sequential_images(filename, path, aug_path):

Marina-Miranovich

import * as tf from '@tensorflow/tfjs';

const model = tf.sequential();
model.add(tf.layers.inputLayer({ inputShape: [1024] }));
model.add(tf.layers.dense({ units: 1024, activation: 'relu' }));
model.add(tf.layers.dense({ units: 1, activation: 'sigmoid' }));
model.compile({
  optimizer: tf.train.adam(1e-6),
  loss: tf.losses.sigmoidCrossEntropy,
  metrics: ['accuracy']

Marina-Miranovich

export const loadModel = async () => {
  const mn = new mobilenet.MobileNet(1, 1);
  mn.path = `file://PATH/TO/model.json`;
  await mn.load();
  return (input): tf.Tensor1D =>
      mn.infer(input, 'global_average_pooling2d_1')
        .reshape([1024]);
};

Marina-Miranovich

const punches = require('fs')
  .readdirSync(Punches)
  .filter(f => f.endsWith('.jpg'))
  .map(f => `${Punches}/${f}`);
const others = require('fs')
  .readdirSync(Others)
  .filter(f => f.endsWith('.jpg'))
  .map(f => `${Others}/${f}`);
const ys = tf.tensor1d(
  new Array(punches.length).fill(1)

Marina-Miranovich

export const readInput = img => imageToInput(readImage(img), TotalChannels);
const readImage = path => jpeg.decode(fs.readFileSync(path), true);
const imageToInput = image => {
  const values = serializeImage(image);
  return tf.tensor3d(values, [image.height, image.width, 3], 'int32');
};
const serializeImage = image => {
  const totalPixels = image.width * image.height;
  const result = new Int32Array(totalPixels * 3);
  for (let i = 0; i < totalPixels; i++) {

Marina-Miranovich

await model.fit(xs, ys, {
  epochs: Epochs,
  batchSize: parseInt(((punches.length + others.length) * BatchSize).toFixed(0)),
  callbacks: {
    onBatchEnd: async (_, logs) => {
      console.log('Cost: %s, accuracy: %s', logs.loss.toFixed(5), logs.acc.toFixed(5));
      await tf.nextFrame();
    }
  }
});

Marina-Miranovich

const video = document.getElementById('cam');
const Layer = 'global_average_pooling2d_1';
const mobilenetInfer = m => (p): tf.Tensor<tf.Rank> => m.infer(p, Layer);
const canvas = document.getElementById('canvas');
const scale = document.getElementById('crop');
const ImageSize = {
  Width: 100,
  Height: 56
};
navigator.mediaDevices

Marina-Miranovich

const grayscale = (canvas: HTMLCanvasElement) => {
  const imageData = canvas.getContext('2d').getImageData(0, 0, canvas.width, canvas.height);
  const data = imageData.data;
  for (let i = 0; i < data.length; i += 4) {
    const avg = (data[i] + data[i + 1] + data[i + 2]) / 3;
    data[i] = avg;
    data[i + 1] = avg;
    data[i + 2] = avg;
  }
  canvas.getContext('2d').putImageData(imageData, 0, 0);

Marina-Miranovich

let mobilenet: (p: any) => tf.Tensor<tf.Rank>;
tf.loadModel('http://localhost:5000/model.json').then(model => {
  mobileNet
    .load()
    .then((mn: any) => mobilenet = mobilenetInfer(mn))
    .then(startInterval(mobilenet, model));
});