-
Notifications
You must be signed in to change notification settings - Fork 1
/
SB_MMoE.py
156 lines (113 loc) · 5.92 KB
/
SB_MMoE.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2019/8/21 4:10
# @Author : Daishijun
# @Site :
# @File : SB_MMoE.py
# @Software : PyCharm
import pandas as pd
from sklearn.preprocessing import LabelEncoder, MinMaxScaler
from sklearn.metrics import accuracy_score, log_loss, roc_auc_score
from sklearn.model_selection import train_test_split
from deepctr.models import xDeepFM, DeepFM, DCN
from deepctr.inputs import SparseFeat, DenseFeat, get_fixlen_feature_names
from tensorflow.python.keras.optimizers import Adam
from deepctr.inputs import input_from_feature_columns, get_linear_logit,build_input_features,combined_dnn_input
from deepctr.layers.core import PredictionLayer, DNN
from deepctr.layers.interaction import FM
from deepctr.layers.utils import concat_fun
from mmoe import MMoE
from tensorflow.python.keras.callbacks import EarlyStopping
import tensorflow as tf
from tensorflow.python.keras.layers import Input
from tensorflow.python.keras.utils import multi_gpu_model
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "2,3"
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
DATA_PATH = '/opt/ByteCamp/'
DATA_FILE = 'bytecamp.data'
data = pd.read_csv(DATA_PATH+DATA_FILE, sep=',')
sparse_features = ['uid', 'u_region_id', 'item_id', 'author_id','music_id']
dense_features = ['duration', 'generate_time']
data[sparse_features] = data[sparse_features].fillna('-1', )
data[dense_features] = data[dense_features].fillna(0,)
target = ['finish', 'like']
# target = ['finish']
data['generate_time'] %= 60 * 60 * 24
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
sparse_feature_columns = [SparseFeat(feat, data[feat].nunique()) #(特征名, 特征不同取值个数)生成SparseFeat对象，name == 特征名，dimension==该特征不同取值个数， dtype ==int32
for feat in sparse_features]
dense_feature_columns = [DenseFeat(feat, 1) #（特征名， dimension==1） 数据dtype == float32
for feat in dense_features]
dnn_feature_columns = sparse_feature_columns + dense_feature_columns
linear_feature_columns = sparse_feature_columns + dense_feature_columns
## 这里有多余的步骤，该方法中间为每个特征设置了Input层，但是没有返回，只返回了特征名称list，其实可以直接从上面的两个list合并得到。
feature_names = get_fixlen_feature_names(linear_feature_columns + dnn_feature_columns)
RIGIONID = 0
train_indexs = data[(data['date'] < 20190708) & (data['u_region_id']==RIGIONID)].index
test_indexs = data[(data['date'] == 20190708) & (data['u_region_id']==RIGIONID)].index
train, test = data.loc[train_indexs], data.loc[test_indexs]
train_model_input = [train[name] for name in feature_names]
test_model_input = [test[name] for name in feature_names]
features = build_input_features(linear_feature_columns + dnn_feature_columns)
inputs_list = list(features.values())
sparse_embedding_list, dense_value_list = input_from_feature_columns(features, dnn_feature_columns,
embedding_size=8,
l2_reg=0.00001, init_std=0.0001,
seed=1024)
dnn_input = combined_dnn_input(sparse_embedding_list,dense_value_list)
# print('test_model_input info')
# print(len(test_model_input))
# print(type(test_model_input[0]))
# print(len(test_model_input[0]))
# print('num_features:',len(test_model_input[0]) )
# MMoE
mmoe_layers = MMoE(units=16, num_experts=8, num_tasks=2)(dnn_input)
print('passed')
output_layers = []
# Build tower layer from MMoE layer
output_info = ['finish', 'like']
for index, task_layer in enumerate(mmoe_layers):
tower_layer = tf.keras.layers.Dense(
units=128,
activation='relu'
)(task_layer)
output_layer1 = tf.keras.layers.Dense(
units=128,
activation='relu'
)(tower_layer)
output_layer = tf.keras.layers.Dense(
units=1,
name=output_info[index],
activation='sigmoid'
)(output_layer1)
output_layers.append(output_layer)
model = tf.keras.Model(inputs=inputs_list, outputs=output_layers)
try:
model = multi_gpu_model(model, gpus=2)
print("Training using multiple GPUs..")
except Exception as e:
print(e)
print("Training using single GPU or CPU..")
model.compile(optimizer=Adam(0.0001), loss={'finish': 'binary_crossentropy', 'like':'binary_crossentropy'},\
metrics=['accuracy','binary_crossentropy'], loss_weights={'finish':0.6, 'like':0.4})
model.summary()
# class_wights = [{0:0.5, 1:0.5}, {0:0.6,1:0.4}]
history = model.fit(x=train_model_input, y={'finish':train['finish'].values, 'like':train['like'].values},
validation_split=0.3,callbacks=[EarlyStopping(monitor='val_loss', patience=1, verbose=0, mode='auto')],\
batch_size=4096, epochs=20, verbose=1)
pred_ans_finish, pred_ans_like = model.predict(test_model_input, batch_size=2**14)
pred_finish = (pred_ans_finish*2).astype(int)
pred_like = (pred_ans_like*2).astype(int)
print('test accuracy ==> finish:{} \t like:{}'.format(round(accuracy_score(test['finish'].values, pred_finish), 4),\
round(accuracy_score(test['like'].values, pred_like), 4)))
print('test LogLoss ==> finish:{} \t like:{}'.format(round(log_loss(test['finish'].values, pred_ans_finish), 4),\
round(log_loss(test['like'].values, pred_ans_like), 4)))
print('test AUC ==> finish:{} \t like:{}'.format(round(roc_auc_score(test['finish'].values, pred_ans_finish), 4),\
round(roc_auc_score(test['like'].values, pred_ans_like), 4)))