Hands-On Machine Learning with TensorFlow.js

Fourier transformation for composite curves

Let's try to decompose a curve that has been constructed from multiple cosine curves. The curve is formed by the addition of two cosine curves:

The first one is an original curve, while the second one is a doubled frequency that's been multiplied by a smaller weight value, 0.3:

const pi = tf.scalar(2.0 * Math.PI);// Original Frequencyconst xs = tf.mul(pi, tf.range(-1.5, 1.5, 0.05));// For double Frequencyconst xs2 = tf.mul(pi, tf.range(-1.5, 1.5, 0.05)).mul(2.0);const ys = tf.cos(xs).add(tf.cos(xs2).mul(0.2));

We expect there to be two peaked frequencies, where one is smaller than the other if we ...