Collaborative filtering model architecture for movie recommendation.

dot_product_bias.py

from fastai import *
from fastbook import *

def create_params(size):
    """
      Pass tensor shape
      Returns normalised model parameters
    """
    return nn.Parameter(torch.zeros(*size).normal_(0, 0.01))

class DotProductBias(Module):
    """
    Model architecture for collaborative filtering
    """
    def __init__(self, n_users, n_movies, n_factors, y_range=(0, 5.5)):
        """
        Initialises model with parameters

        :param n_users: number of users
        :param n_movies: number of movies
        :param n_factors: number of factors
        :param y_range: sigmoid limit
        """
        self.user_factors = create_params([n_users, n_factors])
        self.user_bias = create_params([n_users])
        self.movie_factors = create_params([n_movies, n_factors])
        self.movie_bias = create_params([n_movies])
        self.y_range = y_range
        
    def forward(self, x):
        """
        Applies a forward pass on the dataset passed
        :param x: data as DataLoaders obj
        :return: predictions in sigmoid range (tensor)
        """
        users = self.user_factors[x[:, 0]]
        movies = self.movie_factors[x[:, 1]]
        res = (users*movies).sum(dim=1)
        res += self.user_bias[x[:, 0]] + self.movie_bias[x[:, 1]]
        return sigmoid_range(res, *self.y_range)