Now, let's look at the steps to implement the PCA algorithm using SVD. Consider the training set, , with m data samples. For this dataset, we will progress with the steps as follows:
- Mean normalization: Deduct the mean value of all the data points from the individual data point. With this, we increase the efficiency of model training and get a better error surface shape, . Replace each x(j) with (x(j) - μ(j)).
- Feature scaling: If the different features have different scales, if x1 is the size of a house and x2 is ...