Training the SVD-based model

We're now ready to write our functions that factor our training dataset and build our recommender model. You can see the required functions in this recipe.

How to do it...

We construct the following functions to train our model. Note that these functions are not part of the Recommender class:

In [27]: def initialize(R, K):
...: """
...: Returns initial matrices for an N X M matrix, R and K features.
...:
...: :param R: the matrix to be factorized
...: :param K: the number of latent features
...:
...: :returns: P, Q initial matrices of N x K and M x K sizes
...: """
...: N, M = R.shape
...: P = np.random.rand(N,K)
...: Q = np.random.rand(M,K)
...:
...: return P, Q
  
In [28]: def factor(R, P=None, Q=None, K=2, steps=5000, alpha=0.0002, beta=0.02):
...: """
...: Performs matrix factorization on R with given parameters.
...:
...: :param R: A matrix to be factorized, dimension N x M
...: :param P: an initial matrix of dimension N x K
...: :param Q: an initial matrix...