Method _embedding_input_layer

modelzoo/din/train.py:259–349

(self)

            return embedding

    def _embedding_input_layer(self):
        for key in SEQ_COLUMNS:
            self._feature[key] = tf.strings.split(self._feature[key], '')
            self._feature[key] = tf.sparse.slice(
                self._feature[key], [0, 0], [self._batch_size, MAX_SEQ_LENGTH])

        # get uid embeddings
        if self._adaptive_emb and not self.tf:
            '''Adaptive Embedding Feature Part 1 of 2'''
            adaptive_mask_tensors = {
                'UID': tf.ones([args.batch_size], tf.int32)
            }
            uid_emb = tf.feature_column.input_layer(
                self._feature,
                self._uid_emb_column,
                adaptive_mask_tensors=adaptive_mask_tensors)
        else:
            uid_emb = tf.feature_column.input_layer(self._feature,
                                                    self._uid_emb_column)
        # get embeddings of item and category
        # create embedding table
        if self._ev and not self.tf:
            '''Embedding Variable Feature with get embedding variable API'''
            item_embedding_var = tf.get_embedding_variable(
                'item_embedding_var',
                self._embedding_dim,
                ev_option=self._ev_opt)
            category_embedding_var = tf.get_embedding_variable(
                'category_embedding_var',
                self._embedding_dim,
                ev_option=self._ev_opt)
        elif self._multihash and not self.tf:
            '''Multi-Hash Variable'''
            item_embedding_var = tf.get_multihash_variable(
                'item_embedding_var',
                [[
                    int(self._item_cate_column._num_buckets**0.5),
                    self._embedding_dim
                ],
                 [
                     int(self._item_cate_column._num_buckets /
                         int(self._item_cate_column._num_buckets**0.5)),
                     self._embedding_dim
                 ]])
            category_embedding_var = tf.get_multihash_variable(
                'category_embedding_var',
                [[
                    int(self._category_cate_column._num_buckets**0.5),
                    self._embedding_dim
                ],
                 [
                     int(self._category_cate_column._num_buckets /
                         int(self._category_cate_column._num_buckets**0.5)),
                     self._embedding_dim
                 ]])
        else:
            item_embedding_var = tf.get_variable(
                'item_embedding_var',