Hands-On Machine Learning with TensorFlow.js

The variance of projected datasets

Comparing the variance of the data can demonstrate that PCA is able to achieve the most variance if we convert it into a low-dimensional space. The variance of the data is the Euclidean distance of the difference between the data points and the mean of the feature. As we saw previously, the mean of each feature is calculated by xs.mean(0):

function variance(xs: tf.Tensor) {  const v = xs.sub(xs.mean(0)).pow(2).mean();  console.log(v.dataSync());}

Let's see how the variance can be changed by projecting the data in four ways:

• Projection with 0, 1st features
• Projection with 0, 2nd features
• Projection with 1, 2nd features
• Projection with PCA

This can be done using the following code:

console.log("Variance of ...