def __init__(self, batch_size=256):
self.batch_size = batch_size
self.best_loss = 9999999
self.losses = {'train': [], 'test': []}
self.ub = UserBehavior()
user_feature, item_feature = self.get_inputs()
# 获得特征
self.user_embedding, self.item_feature_embedding = self.ub.load_train_vector(
)
self.user_embedding = tf.keras.layers.Reshape([40],
name='user_embedding')(
self.user_embedding)
self.item_feature_embedding = tf.keras.layers.Reshape(
[43], name="item_feature_embedding")(self.item_feature_embedding)
inference = tf.keras.layers.Lambda(
lambda layer: tf.reduce_sum(layer[0] * layer[1], axis=1),
name="inference")(
(self.user_embedding, self.item_feature_embedding))
inference = tf.keras.layers.Lambda(
lambda layer: tf.expand_dims(layer, axis=1))(inference)
self.model = tf.keras.Model(inputs=[user_feature, item_feature],
outputs=[inference])
self.model.summary()
self.optimizer = tf.keras.optimizers.Adam(lr=0.001)
# MSE损失，将计算值回归到评分
self.ComputeLoss = tf.keras.losses.MeanSquaredError()
self.ComputeMetrics = tf.keras.metrics.MeanAbsoluteError()
if tf.io.gfile.exists(self.MODEL_DIR):
# print('Removing existing model dir: {}'.format(MODEL_DIR))
# tf.io.gfile.rmtree(MODEL_DIR)
pass
else:
tf.io.gfile.makedirs(self.MODEL_DIR)
train_dir = os.path.join(self.MODEL_DIR, 'summaries', 'train')
test_dir = os.path.join(self.MODEL_DIR, 'summaries', 'eval')
checkpoint_dir = os.path.join(self.MODEL_DIR, 'checkpoints')
self.checkpoint_prefix = os.path.join(checkpoint_dir, 'ckpt')
self.checkpoint = tf.train.Checkpoint(model=self.model,
optimizer=self.optimizer)
# Restore variables on creation if a checkpoint exists.
self.checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("finish init")
if avg_loss.result() < best_loss:
best_loss = avg_loss.result()
print("best loss = {}".format(best_loss))
network.checkpoint.save(config.checkpoint_prefix)
def forward(self, xs):
predictions = self.model(xs)
# logits = tf.nn.softmax(predictions)
return predictions
if __name__ == '__main__':
config = RecommenderNetworkConfig()
ub = UserBehavior()
best_loss = 9999999
losses = {'train': [], 'test': []}
network = RecommenderNetwork(config)
# 获得特征
# user_embedding, item_feature_embedding, user_item_score = ub.load_train_vector()
user_embedding_chunks, item_feature_embedding_chunks, user_item_score_chunks = ub.chunk_load_train_vector(
)
# model_input_x = []
# model_input_x.append(user_embedding)
# model_input_x.append(item_feature_embedding)
# model_input_x = list(zip(user_embedding, item_feature_embedding))
model_input_x = list(
zip(user_embedding_chunks, item_feature_embedding_chunks))