-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathfeature_process.py
112 lines (84 loc) · 4.84 KB
/
feature_process.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
#!/usr/bin/env python
# coding: utf-8
# author: stefan 2022-03-24
"""
rank 模型特征处理
"""
from layers.model_layers import ActivationSumPoolingFromDIN, DeepCrossLayer
from layers.tool_layers import *
def build_rank_dumn_feature_columns(city_dict, shangquan_dict, comm_dict, price_dict, area_dict):
# ************************************************************************
# 特征列定义
# ************************************************************************
user_city_seq = tf.keras.Input(shape=(5,), name='user_city_seq', dtype=tf.int64)
user_shangquan_seq = tf.keras.Input(shape=(5,), name='user_shangquan_seq', dtype=tf.int64)
user_comm_seq = tf.keras.Input(shape=(5,), name='user_comm_seq', dtype=tf.int64)
user_price_seq = tf.keras.Input(shape=(5,), name='user_price_seq', dtype=tf.int64)
user_area_seq = tf.keras.Input(shape=(5,), name='user_area_seq', dtype=tf.int64)
item_user1_city_seq = tf.keras.Input(shape=(5,), name='item_user1_city_seq', dtype=tf.int64)
item_user1_shangquan_seq = tf.keras.Input(shape=(5,), name='item_user1_shangquan_seq', dtype=tf.int64)
item_user1_comm_seq = tf.keras.Input(shape=(5,), name='item_user1_comm_seq', dtype=tf.int64)
item_user1_price_seq = tf.keras.Input(shape=(5,), name='item_user1_price_seq', dtype=tf.int64)
item_user1_area_seq = tf.keras.Input(shape=(5,), name='item_user1_area_seq', dtype=tf.int64)
...
# common emb 区域类特征在底层共享
city_Embedding = Embedding(input_dim=400, output_dim=16, mask_zero=False, name="city_emb")
shangquan_Embedding = Embedding(input_dim=15000, output_dim=32, mask_zero=False, name="shangquan_emb")
comm_Embedding = Embedding(input_dim=400000, output_dim=32, mask_zero=False, name="comm_emb")
price_Embedding = Embedding(input_dim=50, output_dim=4, mask_zero=False, name="price_emb")
area_Embedding = Embedding(input_dim=50, output_dim=4, mask_zero=False, name="area_emb")
# ************************************************************************
# 特征分类： 序列embedding, 离散embedding
# ************************************************************************
# 1. embedding ***********************************************************
user_city_id_token = VocabLayer(city_dict, 'city_token')(user_city_seq)
user_city_emb_seq = city_Embedding(user_city_id_token) # 以city_id为index取emb shape(None, 5, emb_size)
user_city_emb = GlobalAveragePooling1D()(user_city_emb_seq) # shape(None, emb_size)
...
# em: candidate item embedding
em = concatenate([item_city_emb, item_shangquan_emb, item_comm_emb, item_price_emb, item_area_emb],
axis=-1, name='em')
# eu: target_user embedding
eu = concatenate([user_city_emb, user_shangquan_emb, user_comm_emb, user_price_emb, user_area_emb],
axis=-1, name='eu')
# Xu target_user history embedding
Xu = getXulEmbedding(user_city_emb_seq, user_shangquan_emb_seq, user_comm_emb_seq, user_price_emb_seq,
user_area_emb_seq, 'Xu')
# # item -> target_user history attention
# ru_ = ActivationSumPoolingFromDIN()([Xu, em])
#
# # r_u: target user representation
# ru = concatenate([ru_, eu], axis=-1, name='ru')
eu1 = concatenate([...], name='eu1')
Xu1 = getXulEmbedding(..., 'Xu1')
eu2 =
Xu2 =
eu3 =
Xu3 =
# context feature
# cross tower
category_features = concatenate([...], axis=1, name='category_features')
dcn_features = DeepCrossLayer(2, category_features.shape[-1], name='dcn_features')(category_features)
# 2.连续特征 ************************************************************************
continue_inputs = [...]
continue_features = concatenate([...], axis=1, name='continue_features')
total_inputs = embedding_inputs + continue_inputs
result = {'total_inputs': total_inputs,
'continue_features': continue_features,
'cross_features': dcn_features,
'em': em, 'eu': eu, 'Xu': Xu, 'eu1': eu1, 'Xu1': Xu1, 'eu2': eu2, 'Xu2': Xu2, 'eu3': eu3, 'Xu3': Xu3
}
return result
def getXulEmbedding(user_city_emb_seq, user_shangquan_emb_seq, user_comm_emb_seq, user_price_emb_seq,
user_area_emb_seq, tag):
Xu = []
for i in range(5):
city_emb = tf.slice(user_city_emb_seq, [0, i, 0], [-1, 1, -1])
shangquan_emb = tf.slice(user_shangquan_emb_seq, [0, i, 0], [-1, 1, -1])
comm_emb = tf.slice(user_comm_emb_seq, [0, i, 0], [-1, 1, -1])
price_emb = tf.slice(user_price_emb_seq, [0, i, 0], [-1, 1, -1])
area_emb = tf.slice(user_area_emb_seq, [0, i, 0], [-1, 1, -1])
item_emb = concatenate([city_emb, shangquan_emb, comm_emb, price_emb, area_emb], axis=-1)
Xu.append(item_emb)
Xu = concatenate(Xu, axis=-2, name=tag + '_emb')
return Xu