AI and Machine Learning for Coders

Chapter 17. Reusing and Converting Python Models to JavaScript

While training in the browser is a powerful option, you may not always want to do this because of the time involved. As you saw in Chapters 15 and 16, even training simple models can lock up the browser for some time. Having a visualization of the progress helped, but it still wasn’t the best of experiences. There are three alternatives to this approach. The first is to train models in Python and convert them to JavaScript. The second is to use existing models that were trained elsewhere and are provided in a JavaScript-ready format. The third is to use transfer learning, introduced in Chapter 3. In that case, features, weights, or biases that have been learned in one scenario can be transferred to another, instead of doing time-consuming relearning. We’ll cover the first two cases in this chapter, and then in Chapter 18 you’ll see how to do transfer learning in JavaScript.

Converting Python-Based Models to JavaScript

Models that have been trained using TensorFlow may be converted to JavaScript using the Python-based tensorflowjs tools. You can install these using:

!pip install tensorflowjs

For example, consider the following simple model that we’ve been using throughout the book:

import numpy as np
import tensorflow as tf
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense

l0 = Dense(units=1, input_shape=[1])
model = Sequential([l0])
model.compile(optimizer='sgd', loss='mean_squared_error ...