forked from shenweichen/DeepCTR
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdeepfm.py
237 lines (197 loc) · 10.3 KB
/
deepfm.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
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
# -*- coding:utf-8 -*-
"""
@author: shenweichen,[email protected]
A keras implementation of DeepFM
Reference:
[1] DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
(https://arxiv.org/abs/1703.04247)
"""
import keras.backend as K
from keras.layers import Input, Dense, Embedding, Concatenate, Activation, Lambda, Reshape, Flatten, Dropout, add, \
subtract
from keras.models import Model
from keras.initializers import RandomNormal, TruncatedNormal,Zeros
from keras.layers import multiply
from keras.engine.topology import Layer
from keras.regularizers import l2
class FMLayer(Layer):
def __init__(self, **kwargs):
super(FMLayer, self).__init__(**kwargs)
def build(self, input_shape):
# Create a trainable weight variable for this layer.
super(FMLayer, self).build(input_shape) # Be sure to call this somewhere!
def call(self, concated_embeds_value):
"""
:param concated_embeds_value: None * field_size * embedding_size
:return: None*1
"""
temp_a = Lambda(lambda x: K.sum(x, axis=1, keepdims=True), )(concated_embeds_value)
temp_b = multiply([concated_embeds_value, concated_embeds_value])
temp_b = Lambda(lambda x: K.sum(x, axis=1, keepdims=True))(temp_b)
cross_term = subtract([temp_a, temp_b])
cross_term = Lambda(lambda x: 0.5 * K.sum(x, axis=2, keepdims=False))(cross_term)
return cross_term
def compute_output_shape(self, input_shape):
return (input_shape[0], 1)
class OutputLayer(Layer):
def __init__(self,activation='sigmoid', **kwargs):
self.activation = activation
super(OutputLayer, self).__init__(**kwargs)
def build(self, input_shape):
# Create a trainable weight variable for this layer.
self.global_bias = self.add_weight(shape=(1,),initializer='zeros',name="global_bias")
super(OutputLayer, self).build(input_shape) # Be sure to call this somewhere!
def call(self, x,):
"""
:param x: None * X
:return: None * X
"""
output = Activation(self.activation)(K.bias_add(x,self.global_bias,data_format='channels_last'))
return output
def compute_output_shape(self, input_shape):
return input_shape
class DeepFM():
def __init__(self, feature_dim_dict, embedding_size=4,
use_fm=True, hidden_size=[], l2_reg_linear=0.00002, l2_reg_fm=0.00002, l2_reg_deep=0.00002,
init_std=0.0001, seed=1024, keep_prob=0.5,final_activation='sigmoid',deep_input_mode='concat',
checkpoint_path=None, bias_feature_dict={'sparse':{},'dense':[]}):
"""
:param feature_dim_dict:
:param embedding_size:
:param use_fm:
:param hidden_size:
:param l2_reg_linear:
:param l2_reg_fm:
:param l2_reg_deep:
:param init_std:
:param seed:
:param keep_prob:
:param final_activation:
:param deep_input_mode:
:param checkpoint_path:
:param bias_feature_dict:
"""
if not isinstance(feature_dim_dict,
dict) or "sparse" not in feature_dim_dict or "dense" not in feature_dim_dict:
raise ValueError(
"feature_dim must be a dict like {'sparse':{'field_1':4,'field_2':3,'field_3':2},'dense':['field_5',]}")
#self.field_dim = field_dim
self.feature_dim = feature_dim_dict
self.embedding_size = embedding_size
self.use_fm = use_fm
self.hidden_size = hidden_size
self.l2_reg_linear = l2_reg_linear
self.l2_reg_fm = l2_reg_fm
self.l2_reg_deep = l2_reg_deep
self.init_std = init_std
self.seed = seed
self.keep_prob = keep_prob
self.activation = "relu"
self.final_activation = final_activation
self.bias_feature_dim = bias_feature_dict
self.checkpoint_path = checkpoint_path
self.deep_input_mode = deep_input_mode
self.model = self.create_model()
def get_model(self, ):
return self.model
def create_model(self, ):
cate_input, continuous_input, bias_cate_input, bias_continuous_input = self.get_input ()
cate_embedding, linear_embedding, bias_cate_embedding = self.create_cate_embedding()
embed_list = [cate_embedding[i](cate_input[i]) for i in range(len(cate_input))]
bias_embed_list = [bias_cate_embedding[i](bias_cate_input[i]) for i in range(len(bias_cate_input))]
linear_term = add([linear_embedding[i](cate_input[i]) for i in range(len(cate_input))])
fm_input = Concatenate(axis=1)(embed_list)
deep_input = Flatten()(fm_input)
if len(continuous_input) > 0:
if len(continuous_input) == 1:
continuous_list = continuous_input[0]
else:
continuous_list = Concatenate()(continuous_input)
deep_input = Concatenate()([deep_input, continuous_list])
fm_out = FMLayer()(fm_input)
deep_out = self.deep_layer(deep_input, self.hidden_size, self.activation, self.keep_prob,
)
if len(self.hidden_size) ==0 and self.use_fm == False:#only linear
final_logit = linear_term,
elif len(self.hidden_size) ==0 and self.use_fm == True:# linear + FM
final_logit = add([linear_term,fm_out])
elif len(self.hidden_size)>0 and self.use_fm == False:# linear + Deep
final_logit = add([linear_term,deep_out])
elif len(self.hidden_size) >0 and self.use_fm == True:# linear + FM + Deep
final_logit = add([linear_term,fm_out,deep_out])
else:
raise NotImplementedError
self.bias_continuous_dim = len(self.bias_feature_dim["dense"])
self.bias_cate_dim = len(self.bias_feature_dim["sparse"])
if self.bias_continuous_dim>0:
bias_continuous_out = Dense(1, activation=None, )(bias_continuous_input)
final_logit = add([final_logit,bias_continuous_out])
if self.bias_cate_dim> 0:
if self.bias_cate_dim == 1:
bias_cate_out = Lambda(lambda x: x)(bias_embed_list) # add(bias_embed_list)
if self.bias_cate_dim > 1:
bias_cate_out = add(bias_embed_list)
final_logit = add([final_logit, bias_cate_out])
output = OutputLayer(self.final_activation)(final_logit)#Activation('sigmoid', name="final_activation", )(final_logit)
output = Reshape([1])(output)
model = Model(inputs=cate_input + continuous_input + bias_cate_input + bias_continuous_input, outputs=output)
return model
def get_input(self, ):
cate_input = [Input(shape=(1,), name='cate' + str(i)+'-'+feat) for i,feat in enumerate(self.feature_dim["sparse"])]
continuous_input = [Input(shape=(1,), name='continuous' + str(i)+'-'+feat) for i,feat in enumerate(self.feature_dim["dense"])]
bias_cate_input = [Input(shape=(1,), name='bias_cate' + str(i)+'-'+feat) for i,feat in enumerate(self.bias_feature_dim["sparse"])]
bias_continuous_input = [Input(shape=(1,), name='bias_continuous' + str(i)+'-'+feat) for i,feat in
enumerate(self.bias_feature_dim["dense"])]
return cate_input, continuous_input, bias_cate_input, bias_continuous_input,
def create_cate_embedding(self,):
"""
:param field_dim:
:param feature_dim:
:param embedding_size:
:param init_std:
:return:
"""
cate_embedding = [Embedding(self.feature_dim["sparse"][feat], self.embedding_size,
embeddings_initializer=RandomNormal(mean=0.0, stddev=self.init_std, seed=self.seed), embeddings_regularizer=l2(self.l2_reg_fm),
name='embed_cate' + str(i)+'-'+feat) for i,feat in
enumerate(self.feature_dim["sparse"])]
linear_embedding = [Embedding(self.feature_dim["sparse"][feat], 1,
embeddings_initializer=RandomNormal(mean=0.0, stddev=self.init_std, seed=self.seed)
, embeddings_regularizer=l2(self.l2_reg_linear), name='embed_linear' + str(i)) for
i,feat in enumerate(self.feature_dim["sparse"])]
bias_cate_embedding = [Embedding(self.bias_feature_dim["sparse"][feat], 1,
embeddings_initializer=RandomNormal(mean=0.0, stddev=self.init_std, seed=self.seed)
, embeddings_regularizer=l2(self.l2_reg_linear), name='embed_bias' + str(i))
for
i,feat in enumerate(self.bias_feature_dim["sparse"])]
return cate_embedding, linear_embedding, bias_cate_embedding#,global_bias
"""
def fm_layer(self, concated_embeds_value):
:param concated_embeds_value: batch_size * feature_dim * feild_dim
:return:
temp_a = Lambda(lambda x: K.sum(x, axis=1, keepdims=True), )(concated_embeds_value)
temp_b = Multiply()([concated_embeds_value, concated_embeds_value])
temp_b = Lambda(lambda x: K.sum(x, axis=1, keepdims=True))(temp_b)
cross_term = subtract([temp_a, temp_b])
cross_term = Lambda(lambda x: 0.5 * K.sum(x, axis=2, keepdims=False))(cross_term)
return cross_term
"""
def deep_layer(self, flatten_embeds_value, hidden_size, activation, keep_prob,):
"""
:param flatten_embeds_value: batch_size * (feature_dim * feild_dim)
:return:
"""
deep_input = flatten_embeds_value
for l in range(len(hidden_size)):
fc = Dense(hidden_size[l], activation=activation, \
kernel_initializer=TruncatedNormal(mean=0.0, stddev=self.init_std, seed=self.seed), \
kernel_regularizer=l2(self.l2_reg_deep))(deep_input)
# if l < len(hidden_size) - 1:
fc = Dropout(1 - keep_prob)(fc)
deep_input = fc
deep_out = Dense(1, activation=None)(deep_input)
return deep_out
if __name__ == "__main__":
model = DeepFM({'sparse':{'field_1':4,'field_2':3,'field_3':2},'dense':['field_5',]}, embedding_size=4, use_fm=True, hidden_size=[4, 4, 4], keep_prob=0.6,).model
model.compile('adam', 'binary_crossentropy', metrics=['binary_crossentropy'])
print("DeepFM compile done")