forked from guoday/Tencent2019_Preliminary_Rank1st
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
165 lines (147 loc) · 6.93 KB
/
train.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
import numpy as np
import pandas as pd
import ctrNet
import tensorflow as tf
from src import misc_utils as utils
import os
import gc
from sklearn import metrics
from sklearn import preprocessing
import random
np.random.seed(2019)
####################################################################################
#单值特征，直接embedding:支持任何可以转换成字符串的数据类型，比如浮点数，会转换成字符串
single_features=['periods_cont','aid','gender','crowd_direction', 'delivery_periods','advertiser', 'good_id', 'good_type', 'ad_type_id','consuptionAbility', 'os', 'work','connectionType','ad_size', 'good_id_advertiser_count', 'good_id_aid_count', 'good_id_ad_size_count', 'good_id_ad_type_id_count', 'good_id_good_id_size', 'advertiser_good_id_count', 'advertiser_aid_count', 'advertiser_ad_size_count', 'advertiser_ad_type_id_count', 'advertiser_good_type_count', 'advertiser_advertiser_size',]
#交叉特征，会使用分解机提取特征:支持任何可以转换成字符串的数据类型。比如浮点数，会转换成字符串
cross_features=[ 'aid','gender','crowd_direction', 'delivery_periods','advertiser', 'good_id', 'good_type', 'ad_type_id','consuptionAbility', 'os','work','connectionType','ad_size']
#多值特征，会使用分解机提取特征:支持字符串数据类型，用空格隔开
multi_features=['aid_uids','age','area','status','behavior','good_id_advertisers','good_id_request_days', 'good_id_positions', 'good_id_period_ids','good_id_wdays','advertiser_good_ids', 'advertiser_request_days', 'advertiser_positions', 'advertiser_period_ids', 'advertiser_wdays']
#稠密特征，直接放入MLP中:主要用于embedding特征，转化率等
dense_features=['uid_w2v_embedding_aid_64_'+str(i) for i in range(64)]+['uid_w2v_embedding_good_id_64_'+str(i) for i in range(64)]+['uid_w2v_embedding_advertiser_64_'+str(i) for i in range(64)]
dense_features+=['uid_aid_aid_deepwalk_embedding_64_'+str(i) for i in range(64)]+['uid_good_id_good_id_deepwalk_embedding_64_'+str(i) for i in range(64)]
dense_features+=['periods_on_'+str(i) for i in range(48)]
#key-values 特征，将稠密特征转换成向量: 浮点数类型，且数值在[0,1]之间
kv_features=['history_aid_imp', 'history_aid_bid', 'history_aid_pctr', 'history_aid_quality_ecpm', 'history_aid_totalEcpm', 'good_id_advertiser_count', 'good_id_aid_count', 'good_id_ad_size_count', 'good_id_ad_type_id_count', 'good_id_good_id_size', 'advertiser_good_id_count','advertiser_aid_count', 'advertiser_ad_size_count', 'advertiser_ad_type_id_count', 'advertiser_good_type_count', 'advertiser_advertiser_size', 'good_id_imp_median', 'good_id_imp_std', 'good_id_imp_min', 'good_id_imp_max', 'advertiser_imp_mean', 'advertiser_imp_median', 'advertiser_imp_std', 'advertiser_imp_min', 'advertiser_imp_max','create_timestamp']
####################################################################################
#参数
hparam=tf.contrib.training.HParams(
model='CIN',
norm=True,
batch_norm_decay=0.9,
hidden_size=[1024,512],
dense_hidden_size=[300],
cross_layer_sizes=[128,128],
k=16,
single_k=16,
max_length=100,
cross_hash_num=int(5e6),
single_hash_num=int(5e6),
multi_hash_num=int(1e6),
batch_size=32,
infer_batch_size=2**14,
optimizer="adam",
dropout=0,
kv_batch_num=20,
learning_rate=0.0002,
num_display_steps=1000,
num_eval_steps=1000,
epoch=1, #don't modify
metric='SMAPE',
activation=['relu','relu','relu'],
init_method='tnormal',
cross_activation='relu',
init_value=0.001,
single_features=single_features,
cross_features=cross_features,
multi_features=multi_features,
dense_features=dense_features,
kv_features=kv_features,
label='imp',
model_name="CIN")
utils.print_hparams(hparam)
####################################################################################
#读取数据
test=pd.read_pickle('data/test_NN.pkl')
dev=pd.read_pickle('data/dev_NN.pkl')
train=pd.read_pickle('data/train_NN_0.pkl')
train_dev=pd.read_pickle('data/train_dev_NN_0.pkl')
train['gold_imp']=train['imp']
dev['gold_imp']=dev['imp']
train['imp']=train['imp'].apply(lambda x:np.log(x+1))
train_dev['imp']=train_dev['imp'].apply(lambda x:np.log(x+1))
####################################################################################
#检验模型
print(dev.shape)
print(train_dev.shape)
scaler=preprocessing.MinMaxScaler(feature_range=(0,8))
scaler.fit(train_dev[['imp']])
hparam.train_scaler=scaler
hparam.test_scaler=scaler
print("*"*80)
model=ctrNet.build_model(hparam)
model.train(train_dev,dev)
dev_preds=np.zeros(len(dev))
dev_preds=model.infer(dev)
dev_preds=np.exp(dev_preds)-1
print(np.mean(dev_preds))
print("*"*80)
####################################################################################
#测试模型
print(test.shape)
print(train.shape)
scaler=preprocessing.MinMaxScaler(feature_range=(0,8))
scaler.fit(train[['imp']])
hparam.train_scaler=scaler
hparam.test_scaler=scaler
index=set(range(train.shape[0]))
K_fold=[]
for i in range(5):
if i == 4:
tmp=index
else:
tmp=random.sample(index,int(1.0/5*train.shape[0]))
index=index-set(tmp)
print("Number:",len(tmp))
K_fold.append(tmp)
train_preds=np.zeros(len(train))
test_preds=np.zeros(len(test))
scores=[]
train['gold']=True
for i in range(5):
print("Fold",i)
dev_index=K_fold[i]
train_index=[]
for j in range(5):
if j!=i:
train_index+=K_fold[j]
for k in range(2):
model=ctrNet.build_model(hparam)
score=model.train(train.loc[train_index],train.loc[dev_index])
scores.append(score)
train_preds[list(dev_index)]+=model.infer(train.loc[list(dev_index)])/2
test_preds+=model.infer(test)/10
print(np.mean((np.exp(test_preds*10/(i*2+k+1))-1)))
try:
del model
gc.collect()
except:
pass
train_preds=np.exp(train_preds)-1
test_preds=np.exp(test_preds)-1
####################################################################################
#输出
print(scores)
print(np.mean(scores))
print(train_preds.mean())
print(dev_preds.mean())
print(test_preds.mean())
train_fea = train[['aid','request_day']]
train_fea['nn_preds']=train_preds
dev['nn_preds'] = dev_preds
dev_fea=dev[['aid','bid','gold','imp','nn_preds']]
test['nn_preds'] = test_preds
test_fea=test[['aid','nn_preds']]
train_fea.to_csv('submission/nn_pred_{}_train.csv'.format(hparam.model_name),index=False)
test_fea.to_csv('submission/nn_pred_{}_test.csv'.format(hparam.model_name),index=False)
dev_fea.to_csv('submission/nn_pred_{}_dev.csv'.format(hparam.model_name),index=False)
####################################################################################