Hands-On Recommendation Systems with Python

Supervised learning and dimensionality reduction

Consider our ratings matrix once again. It is of the m × n shape, where every row represents one of the m users and every column represents one of the n items.

Let's now remove one of the n columns (say n_j). We now have an m × (n-1) matrix. If we treat the m × (n-1) matrix as the predictor variables and n_j as the target variable, we can use supervised learning algorithms to train on the values available in n_j to predict values that are not. This can be repeated n times for every column to eventually complete our matrix.

One big problem is that most supervised learning algorithms do not work with missing data. In standard problems, it is common practice to impute the missing values with the ...