"""
Module description:
"""
__version__ = '0.3.1'
__author__ = 'Vito Walter Anelli, Claudio Pomo, Daniele Malitesta'
__email__ = 'vitowalter.anelli@poliba.it, claudio.pomo@poliba.it, daniele.malitesta@poliba.it'
import tensorflow as tf
from tensorflow import keras
import numpy as np
[docs]class DeepStyleModel(keras.Model):
def __init__(self,
factors=20,
learning_rate=0.001,
l_w=0,
num_image_feature=2048,
num_users=100,
num_items=100,
name="DeepStyle",
random_seed=42,
**kwargs):
super().__init__(name=name, **kwargs)
tf.random.set_seed(random_seed)
self._factors = factors
self._learning_rate = learning_rate
self.l_w = l_w
self._num_image_feature = num_image_feature
self._num_items = num_items
self._num_users = num_users
self.initializer = tf.initializers.GlorotUniform()
self.Gu = tf.Variable(self.initializer(shape=[self._num_users, self._factors]), name='Gu', dtype=tf.float32)
self.Gi = tf.Variable(self.initializer(shape=[self._num_items, self._factors]), name='Gi', dtype=tf.float32)
self.Li = tf.Variable(
self.initializer(shape=[self._num_items, self._factors]),
name='Li', dtype=tf.float32)
self.E = tf.Variable(
self.initializer(shape=[self._num_image_feature, self._factors]),
name='E', dtype=tf.float32)
self.optimizer = tf.optimizers.Adam(self._learning_rate)
[docs] @tf.function
def call(self, inputs, training=None):
user, item, feature_i = inputs
gamma_u = tf.squeeze(tf.nn.embedding_lookup(self.Gu, user))
gamma_i = tf.squeeze(tf.nn.embedding_lookup(self.Gi, item))
l_i = tf.squeeze(tf.nn.embedding_lookup(self.Li, item))
xui = tf.reduce_sum(gamma_u * (tf.matmul(feature_i, self.E) - l_i + gamma_i), 1)
return xui, gamma_u, gamma_i, feature_i, l_i
[docs] @tf.function
def train_step(self, batch):
user, pos, feat_pos, neg, feat_neg = batch
with tf.GradientTape() as t:
# Clean Inference
xu_pos, gamma_u, gamma_pos, _, l_pos = self.call(inputs=(user, pos, feat_pos), training=True)
xu_neg, _, gamma_neg, _, l_neg = self.call(inputs=(user, neg, feat_neg), training=True)
result = tf.clip_by_value(xu_pos - xu_neg, -80.0, 1e8)
loss = tf.reduce_sum(tf.nn.softplus(-result))
# Regularization Component
reg_loss = self.l_w * tf.reduce_sum([tf.nn.l2_loss(gamma_u),
tf.nn.l2_loss(gamma_pos),
tf.nn.l2_loss(gamma_neg),
tf.nn.l2_loss(l_pos),
tf.nn.l2_loss(l_neg)])
# Loss to be optimized
loss += reg_loss
grads = t.gradient(loss, [self.Gu, self.Gi, self.E, self.Li])
self.optimizer.apply_gradients(zip(grads, [self.Gu, self.Gi, self.E, self.Li]))
return loss
[docs] @tf.function
def predict_batch(self, start, stop, gi, li, fi):
return tf.reduce_sum(self.Gu[start:stop] * (tf.matmul(fi, self.E) - li + gi), axis=1)
[docs] @tf.function
def predict_item_batch(self, start, stop, start_item, stop_item, feat):
return tf.matmul(self.Gu[start:stop], (tf.matmul(feat, self.E) - self.Li[start_item:(stop_item + 1)]
+ self.Gi[start_item:(stop_item + 1)]), transpose_b=True)
[docs] @tf.function
def get_config(self):
raise NotImplementedError
[docs] @tf.function
def get_top_k(self, preds, train_mask, k=100):
return tf.nn.top_k(tf.where(train_mask, preds, -np.inf), k=k, sorted=True)