/
ncf.py
101 lines (82 loc) · 6.15 KB
/
ncf.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
from deepctr_torch.inputs import create_embedding_matrix
from deepctr_torch.inputs import SparseFeat, DenseFeat, build_input_features
from deepctr_torch.layers.core import DNN
from .PLBaseModel import PLBaseModel
from ..utils import combined_dnn_input
class NCF(PLBaseModel):
def __init__(self, user_feature_columns, item_feature_columns,
user_gmf_embedding_dim=20, item_gmf_embedding_dim=20,
user_mlp_embedding_dim=20, item_mlp_embedding_dim=20, dnn_use_bn=False,
dnn_hidden_units=[64, 32], dnn_activation='relu', l2_reg_dnn=0,
dnn_dropout=0,
l2_reg_linear=0.00001, l2_reg_embedding=0.00001, init_std=0.0001, seed=1024, task='binary', device='cpu', **kwargs):
super().__init__(user_feature_columns, item_feature_columns, l2_reg_linear=l2_reg_linear, l2_reg_embedding=l2_reg_embedding, init_std=init_std, seed=seed, task=task, device=device, **kwargs)
if len(user_feature_columns) > 1 or len(item_feature_columns) > 1:
raise ValueError("目前 NCF 只支持 UserId 和 ItemId 作为特征")
# 这里的 user_feature_columns 和 dssm 之类的不同，因为 gmf 和 mlp 部分的 embeding 层不一样
self.dnn_hidden_units = dnn_hidden_units
self.dnn_activation = dnn_activation
self.l2_reg_dnn = l2_reg_dnn
self.dnn_dropout = dnn_dropout
self.user_mlp_embedding_dim= user_mlp_embedding_dim
self.item_mlp_embedding_dim = item_mlp_embedding_dim
self.dnn_use_bn = dnn_use_bn
self.init_std = init_std
self.l2_reg_embedding = l2_reg_embedding
self.l2_reg_linear = l2_reg_linear
self.seed = seed
self.task = task
# self.device = device
# 计算模型的复杂度
self.user_gmf_embedding_dim = user_gmf_embedding_dim
self.item_gmf_embedding_dim = item_gmf_embedding_dim
# GMF Part
self.user_gmf_feature_columns = [SparseFeat(feat, vocabulary_size=size,
embedding_dim=self.user_gmf_embedding_dim)
for feat, size, *_ in self.user_feature_columns]
self.item_gmf_feature_columns = [SparseFeat(feat, vocabulary_size=size, embedding_dim=self.item_gmf_embedding_dim)
for feat, size, *_ in self.item_feature_columns]
self.feature_index = build_input_features(self.user_gmf_feature_columns + self.item_gmf_feature_columns)
self.gmf_embedding_dict = create_embedding_matrix(self.user_gmf_feature_columns + self.item_gmf_feature_columns,
self.init_std, linear=False, sparse=False,
device=self.device)
self.user_mlp_feature_columns = [SparseFeat(feat, vocabulary_size=size, embedding_dim=self.user_mlp_embedding_dim)
for feat, size, *_ in self.user_feature_columns]
self.item_mlp_feature_columns = [SparseFeat(feat, vocabulary_size=size, embedding_dim=self.item_mlp_embedding_dim)
for feat, size, *_ in self.item_feature_columns]
self.mlp_embedding_dict = create_embedding_matrix(self.user_mlp_feature_columns + self.item_mlp_feature_columns,
self.init_std, linear=False, sparse=False,
device=self.device)
self.mlp_dnn = DNN(self.compute_input_dim(self.user_mlp_feature_columns + self.item_mlp_feature_columns), self.dnn_hidden_units, self.dnn_activation,
seed = self.seed, device=device)
self.predict_layer = DNN(self.dnn_hidden_units[-1] + item_gmf_embedding_dim, hidden_units=[1], activation='sigmoid',seed=self.seed, device=device)
def forward(self, X):
user_gmf_sparse_embedding_list, user_gmf_dense_value_list = self.input_from_feature_columns(X,
self.user_gmf_feature_columns,
self.gmf_embedding_dict
)
item_gmf_sparse_embedding_list, item_gmf_dense_value_list = self.input_from_feature_columns(X,
self.item_gmf_feature_columns,
self.gmf_embedding_dict)
user_gmf_input = combined_dnn_input(user_gmf_sparse_embedding_list, user_gmf_dense_value_list)
item_gmf_input = combined_dnn_input(item_gmf_sparse_embedding_list, item_gmf_dense_value_list)
# 对应元素相差
gmf_out = user_gmf_input * item_gmf_input
# MLP part
user_mlp_sparse_embedding_list, user_mlp_dense_value_list = self.input_from_feature_columns(X,
self.user_mlp_feature_columns,
self.mlp_embedding_dict)
item_mlp_sparse_embedding_list, item_mlp_dense_value_list = self.input_from_feature_columns(X,
self.item_mlp_feature_columns,
self.mlp_embedding_dict)
user_mlp_input = combined_dnn_input(
user_mlp_sparse_embedding_list, user_mlp_dense_value_list)
item_mlp_input = combined_dnn_input(
item_mlp_sparse_embedding_list, item_mlp_dense_value_list)
mlp_input = combined_dnn_input([user_mlp_input, item_mlp_input])
mlp_out = self.mlp_dnn(mlp_input)
# Fusion of GMF and MLP
neumf_input = combined_dnn_input([gmf_out, mlp_out])
neumf_out = self.predict_layer(neumf_input)
neumf_out = neumf_out.squeeze()
return neumf_out