def _build_embedding(self):
"""The field embedding layer. Initialization of embedding variables."""
hparams = self.hparams
self.user_vocab_length = len(load_dict(hparams.user_vocab))
self.item_vocab_length = len(load_dict(hparams.item_vocab))
self.cate_vocab_length = len(load_dict(hparams.cate_vocab))
self.user_embedding_dim = hparams.user_embedding_dim
self.item_embedding_dim = hparams.item_embedding_dim
self.cate_embedding_dim = hparams.cate_embedding_dim
with tf.variable_scope("embedding", initializer=self.initializer):
self.user_lookup = tf.get_variable(
name="user_embedding",
shape=[self.user_vocab_length, self.user_embedding_dim],
dtype=tf.float32,
)
self.item_lookup = tf.get_variable(
name="item_embedding",
shape=[self.item_vocab_length, self.item_embedding_dim],
dtype=tf.float32,
)
self.cate_lookup = tf.get_variable(
name="cate_embedding",
shape=[self.cate_vocab_length, self.cate_embedding_dim],
dtype=tf.float32,
)
def __init__(self, hparams, graph, col_spliter="\t"):
"""Initialize an iterator. Create necessary placeholders for the model.
Args:
hparams (obj): Global hyper-parameters. Some key settings such as #_feature and #_field are there.
graph (obj): the running graph. All created placeholder will be added to this graph.
col_spliter (str): column spliter in one line.
"""
self.col_spliter = col_spliter
user_vocab, item_vocab, cate_vocab = (
hparams.user_vocab,
hparams.item_vocab,
hparams.cate_vocab,
)
self.userdict, self.itemdict, self.catedict = (
load_dict(user_vocab),
load_dict(item_vocab),
load_dict(cate_vocab),
)
self.max_seq_length = hparams.max_seq_length
self.batch_size = hparams.batch_size
self.iter_data = dict()
self.graph = graph
with self.graph.as_default():
self.labels = tf.placeholder(tf.float32, [None, 1], name="label")
self.users = tf.placeholder(tf.int32, [None], name="users")
self.items = tf.placeholder(tf.int32, [None], name="items")
self.cates = tf.placeholder(tf.int32, [None], name="cates")
self.item_history = tf.placeholder(
tf.int32, [None, self.max_seq_length], name="item_history"
)
self.item_cate_history = tf.placeholder(
tf.int32, [None, self.max_seq_length], name="item_cate_history"
)
self.mask = tf.placeholder(
tf.int32, [None, self.max_seq_length], name="mask"
)
self.time = tf.placeholder(tf.float32, [None], name="time")
self.time_diff = tf.placeholder(
tf.float32, [None, self.max_seq_length], name="time_diff"
)
self.time_from_first_action = tf.placeholder(
tf.float32, [None, self.max_seq_length], name="time_from_first_action"
)
self.time_to_now = tf.placeholder(
tf.float32, [None, self.max_seq_length], name="time_to_now"
)