RecommenderNet.py

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers


class RecommenderNet(keras.Model):
    """Objet de type réseau-neuronal, utilisé en class de type « Model »."""
    def __init__(self, num_users, num_books, embedding_size, **kwargs):
        super(RecommenderNet, self).__init__(**kwargs)
        self.num_users = num_users
        self.num_books = num_books
        self.embedding_size = embedding_size
        self.user_embedding = layers.Embedding(
            num_users,
            embedding_size,
            embeddings_initializer="he_normal",
            embeddings_regularizer=keras.regularizers.l2(1e-6),
        )
        self.user_bias = layers.Embedding(num_users, 1)
        self.book_embedding = layers.Embedding(
            num_books,
            embedding_size,
            embeddings_initializer="he_normal",
            embeddings_regularizer=keras.regularizers.l2(1e-6),
        )
        self.book_bias = layers.Embedding(num_books, 1)

    def call(self, inputs):
        user_vector = self.user_embedding(inputs[:, 0])
        user_bias = self.user_bias(inputs[:, 0])
        book_vector = self.book_embedding(inputs[:, 1])
        book_bias = self.book_bias(inputs[:, 1])
        dot_user_book = tf.tensordot(user_vector, book_vector, 2)
        x = dot_user_book + user_bias + book_bias
        return tf.nn.sigmoid(x)