Machine Learning Algorithms - Second Edition

Whitening

The StandardScaler class operates in a feature-wise fashion, however, sometimes, it's useful to transform the whole dataset so as to force it to have an identity covariance matrix (to improve the performances of many algorithms that are sensitive to the number of independent components):

The goal is to find a transformation matrix A (called the whitening matrix) so that the new dataset X^' = XA^T has an identity covariance C' (we are assuming that X is zero-centered or, alternatively, it has zero mean). The procedure is quite simple (it can be found in Mastering Machine Learning Algorithms, Bonaccorso G., Packt Publishing, 2018), but ...