-
Notifications
You must be signed in to change notification settings - Fork 12
/
feature_engineering.py
272 lines (229 loc) · 13.7 KB
/
feature_engineering.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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
# -*- coding: utf-8 -*-
import pandas as pd
from datetime import datetime
import numpy as np
from pandas.core.frame import DataFrame
"""
构造特征,保存到log_pre.csv
day_of_week_0_sum到day_of_week_6_sum表示用户周一到周日的模块点击次数
将一天划分成多个时间段后，time_of_day_0_sum到time_of_day_6_sum 分别统计用户在各个时间段的模块点击次数
TCH_TYP_0_sum 和 TCH_TYP_2_sum 表示用户对不同事件类型的模块点击次数
click_count表示用户点击的模块总数
EVT_LBL_set_len表示用户点击了多少个独特的模块（跟click_count相比，去掉了重复的模块)
"""
def log_pre():
train_agg = pd.read_csv('train/train_agg.csv', sep='\t')
train_flg = pd.read_csv('train/train_flg.csv', sep='\t')
train_log = pd.read_csv('train/train_log.csv', sep='\t')
test_agg = pd.read_csv('test/test_agg.csv', sep='\t')
test_log = pd.read_csv('test/test_log.csv', sep='\t')
log = pd.concat([train_log, test_log], copy=False)
log['day_of_week'] = log['OCC_TIM'].map(lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S").weekday())
log['day'] = log['OCC_TIM'].map(lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S").day)
## sleep: 0-6 , go to work:7-9, work:10-12, sleep:13-14, work 15-17,dinner 18-20, rest 21-23
time_of_day = [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6]
log['time_of_day'] = log['OCC_TIM'].map(lambda x: time_of_day[datetime.strptime(x, "%Y-%m-%d %H:%M:%S").hour])
dw = pd.get_dummies(log["day_of_week"], prefix="day_of_week")
td = pd.get_dummies(log["time_of_day"], prefix="time_of_day")
tch = pd.get_dummies(log["TCH_TYP"], prefix="TCH_TYP")
log = pd.concat([log, dw, td, tch], axis=1)
log['click_count'] = 1
c0 = log.groupby(['USRID'], as_index=False)['click_count'].agg({'click_count': np.sum})
c1 = log.groupby(['USRID'], as_index=False)['EVT_LBL'].agg({'EVT_LBL_set_len': lambda x: len(set(x))})
log.drop(['EVT_LBL', 'OCC_TIM', 'TCH_TYP', 'day_of_week', 'day', 'time_of_day'], axis=1, inplace=True)
d0 = log.groupby(['USRID'], as_index=False)['day_of_week_0'].agg({'day_of_week_0_sum': np.sum})
d1 = log.groupby(['USRID'], as_index=False)['day_of_week_1'].agg({'day_of_week_1_sum': np.sum})
d2 = log.groupby(['USRID'], as_index=False)['day_of_week_2'].agg({'day_of_week_2_sum': np.sum})
d3 = log.groupby(['USRID'], as_index=False)['day_of_week_3'].agg({'day_of_week_3_sum': np.sum})
d4 = log.groupby(['USRID'], as_index=False)['day_of_week_4'].agg({'day_of_week_4_sum': np.sum})
d5 = log.groupby(['USRID'], as_index=False)['day_of_week_5'].agg({'day_of_week_5_sum': np.sum})
d6 = log.groupby(['USRID'], as_index=False)['day_of_week_6'].agg({'day_of_week_6_sum': np.sum})
t0 = log.groupby(['USRID'], as_index=False)['time_of_day_0'].agg({'time_of_day_0_sum': np.sum})
t1 = log.groupby(['USRID'], as_index=False)['time_of_day_1'].agg({'time_of_day_1_sum': np.sum})
t2 = log.groupby(['USRID'], as_index=False)['time_of_day_2'].agg({'time_of_day_2_sum': np.sum})
t3 = log.groupby(['USRID'], as_index=False)['time_of_day_3'].agg({'time_of_day_3_sum': np.sum})
t4 = log.groupby(['USRID'], as_index=False)['time_of_day_4'].agg({'time_of_day_4_sum': np.sum})
t5 = log.groupby(['USRID'], as_index=False)['time_of_day_5'].agg({'time_of_day_5_sum': np.sum})
t6 = log.groupby(['USRID'], as_index=False)['time_of_day_6'].agg({'time_of_day_6_sum': np.sum})
p0 = log.groupby(['USRID'], as_index=False)['TCH_TYP_0'].agg({'TCH_TYP_0_sum': np.sum})
p2 = log.groupby(['USRID'], as_index=False)['TCH_TYP_2'].agg({'TCH_TYP_2_sum': np.sum})
log_pre = pd.merge(d0, d1, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, d2, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, d3, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, d4, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, d5, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, d6, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t0, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t1, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t2, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t3, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t4, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t5, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, t6, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, p0, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, p2, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, c0, on=['USRID'], how='left', copy=False)
log_pre = pd.merge(log_pre, c1, on=['USRID'], how='left', copy=False)
log_pre.to_csv('log_pre.csv', sep='\t', index=False)
"""
构造特征,保存到log_EVT_square.csv
将用户的点击模块划分成三个级别EVT_LBL_1，EVT_LBL_2，EVT_LBL_3，统计用户在一个级别内对不同模块的点击次数
如特征EVT_1_num_12表示用户对第一个级别的第12个模块的点击次数
特征EVT_2_num_21表示用户对第二个级别的第21个模块的点击次数，依此类推。。。。。。
"""
def log_EVT_square():
train_log = pd.read_csv('train/train_log.csv', sep='\t')
test_log = pd.read_csv('test/test_log.csv', sep='\t')
log = pd.concat([train_log, test_log], copy=False)
log['EVT_LBL_1'] = log['EVT_LBL'].apply(lambda x: x.split('-')[0])
log['EVT_LBL_2'] = log['EVT_LBL'].apply(lambda x: x.split('-')[1])
log['EVT_LBL_3'] = log['EVT_LBL'].apply(lambda x: x.split('-')[2])
full_EVT_1 = list(set(log['EVT_LBL_1']))
full_EVT_2 = list(set(log['EVT_LBL_2']))
full_EVT_3 = list(set(log['EVT_LBL_3']))
def EVT_LBL_1_TONUM(s):
return full_EVT_1.index(s)
def EVT_LBL_2_TONUM(s):
return full_EVT_2.index(s)
def EVT_LBL_3_TONUM(s):
return full_EVT_3.index(s)
log['EVT_LBL_1'] = log.EVT_LBL_1.apply(EVT_LBL_1_TONUM)
log['EVT_LBL_2'] = log.EVT_LBL_2.apply(EVT_LBL_2_TONUM)
log['EVT_LBL_3'] = log.EVT_LBL_3.apply(EVT_LBL_3_TONUM)
Users_set = DataFrame(list(set(log['USRID'])))
Users_set.columns = ['USRID']
###统计每个用户各个level的不同模块的点击次数，i表示full_EVT的第i个元素
t = log[['USRID', 'EVT_LBL_1']][log.EVT_LBL_1 == 0]
t['EVT_1_num_0'] = 1
t = t.groupby(['USRID', 'EVT_LBL_1']).agg('sum').reset_index()
t.drop('EVT_LBL_1', axis=1, inplace=True)
log_EVT_1 = pd.merge(Users_set, t, on='USRID', how='left')
for i in range(1, len(full_EVT_1)):
t = log[['USRID', 'EVT_LBL_1']][log.EVT_LBL_1 == i]
t['EVT_1_num_' + str(i)] = 1
t = t.groupby(['USRID', 'EVT_LBL_1']).agg('sum').reset_index()
t.drop('EVT_LBL_1', axis=1, inplace=True)
log_EVT_1 = pd.merge(log_EVT_1, t, on='USRID', how='left')
for i in range(len(full_EVT_1)):
log_EVT_1['EVT_1_num_' + str(i)][log_EVT_1['EVT_1_num_' + str(i)].isnull()] = 0
t = log[['USRID', 'EVT_LBL_2']][log.EVT_LBL_2 == 0]
t['EVT_2_num_0'] = 1
t = t.groupby(['USRID', 'EVT_LBL_2']).agg('sum').reset_index()
t.drop('EVT_LBL_2', axis=1, inplace=True)
log_EVT_2 = pd.merge(Users_set, t, on='USRID', how='left')
for i in range(1, len(full_EVT_2)):
t = log[['USRID', 'EVT_LBL_2']][log.EVT_LBL_2 == i]
t['EVT_2_num_' + str(i)] = 1
t = t.groupby(['USRID', 'EVT_LBL_2']).agg('sum').reset_index()
t.drop('EVT_LBL_2', axis=1, inplace=True)
log_EVT_2 = pd.merge(log_EVT_2, t, on='USRID', how='left')
for i in range(len(full_EVT_2)):
log_EVT_2['EVT_2_num_' + str(i)][log_EVT_2['EVT_2_num_' + str(i)].isnull()] = 0
t = log[['USRID', 'EVT_LBL_3']][log.EVT_LBL_3 == 0]
t['EVT_3_num_0'] = 1
t = t.groupby(['USRID', 'EVT_LBL_3']).agg('sum').reset_index()
t.drop('EVT_LBL_3', axis=1, inplace=True)
log_EVT_3 = pd.merge(Users_set, t, on='USRID', how='left')
for i in range(1, len(full_EVT_3)):
t = log[['USRID', 'EVT_LBL_3']][log.EVT_LBL_3 == i]
t['EVT_3_num_' + str(i)] = 1
t = t.groupby(['USRID', 'EVT_LBL_3']).agg('sum').reset_index()
t.drop('EVT_LBL_3', axis=1, inplace=True)
log_EVT_3 = pd.merge(log_EVT_3, t, on='USRID', how='left')
for i in range(len(full_EVT_3)):
log_EVT_3['EVT_3_num_' + str(i)][log_EVT_3['EVT_3_num_' + str(i)].isnull()] = 0
log_EVT = pd.merge(log_EVT_1, log_EVT_2, on='USRID', how='left')
log_EVT = pd.merge(log_EVT, log_EVT_3, on='USRID', how='left')
## 对用户点击模块的次数取平方
user = log_EVT.pop('USRID')
log_EVT_square = np.square(log_EVT)
user = DataFrame(user)
log_EVT_square = pd.concat([user, log_EVT_square], axis=1)
log_EVT_square.to_csv('log_EVT_square.csv', sep='\t', index=False)
"""
构造新的用户时间特征，与之前的log_pre.csv，log_EVT_square.csv合并，
形成最终的训练集all_train.csv，最终的测试集test_set.csv
"""
def log_tabel(data):
data['day'] = data['OCC_TIM'].map(lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S").day)
data['hour'] = data['OCC_TIM'].map(lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S").hour)
data['min'] = data['OCC_TIM'].map(lambda x: datetime.strptime(x, "%Y-%m-%d %H:%M:%S").minute)
EVT_LBL_len = data.groupby(by=['USRID'], as_index=False)['EVT_LBL'].agg({'EVT_LBL_len': len})
# 小时均值特征
t0 = data.groupby('USRID')['hour'].mean().reset_index()
t0.columns = ['USRID', 'user_mean_hour']
# 小时var方差特征
t1 = data.groupby('USRID')['hour'].var().reset_index()
t1.columns = ['USRID', 'user_var_hour']
# 天均值特征
t2 = data.groupby('USRID')['day'].mean().reset_index()
t2.columns = ['USRID', 'user_mean_day']
# 天方差特征
t3 = data.groupby('USRID')['day'].var().reset_index()
t3.columns = ['USRID', 'user_var_day']
# 小时min,max,时间差特征
t4 = data.groupby('USRID')['hour'].min().reset_index()
t4.columns = ['USRID', 'user_min_hour']
t5 = data.groupby('USRID')['hour'].max().reset_index()
t5.columns = ['USRID', 'user_max_hour']
diff = t5['user_max_hour'] - t4['user_min_hour']
user = t4['USRID']
t6 = pd.concat([user, diff], axis=1)
t6.columns = ['USRID', 'user_diff_hour']
# 天min,max,时间差特征
t7 = data.groupby('USRID')['day'].min().reset_index()
t7.columns = ['USRID', 'user_min_day']
t8 = data.groupby('USRID')['day'].max().reset_index()
t8.columns = ['USRID', 'user_max_day']
diff2 = t8['user_max_day'] - t7['user_min_day']
user2 = t7['USRID']
t9 = pd.concat([user2, diff2], axis=1)
t9.columns = ['USRID', 'user_diff_day']
return EVT_LBL_len, t0, t1, t2, t3, t4, t5, t6, t7, t8, t9
def Merge():
train_agg = pd.read_csv('train/train_agg.csv', sep='\t')
train_flg = pd.read_csv('train/train_flg.csv', sep='\t')
train_log = pd.read_csv('train/train_log.csv', sep='\t')
all_train = pd.merge(train_flg, train_agg, on=['USRID'], how='left')
EVT_LBL_len, t0, t1, t2, t3, t4, t5, t6, t7, t8, t9 = log_tabel(train_log)
all_train = pd.merge(all_train, EVT_LBL_len, on=['USRID'], how='left')
all_train = pd.merge(all_train, t0, on=['USRID'], how='left')
all_train = pd.merge(all_train, t1, on=['USRID'], how='left')
all_train = pd.merge(all_train, t2, on=['USRID'], how='left')
all_train = pd.merge(all_train, t3, on=['USRID'], how='left')
all_train = pd.merge(all_train, t4, on=['USRID'], how='left')
all_train = pd.merge(all_train, t5, on=['USRID'], how='left')
all_train = pd.merge(all_train, t6, on=['USRID'], how='left')
all_train = pd.merge(all_train, t7, on=['USRID'], how='left')
all_train = pd.merge(all_train, t8, on=['USRID'], how='left')
all_train = pd.merge(all_train, t9, on=['USRID'], how='left')
log_pre = pd.read_csv('log_pre.csv', sep='\t')
all_train = pd.merge(all_train, log_pre, on=['USRID'], how='left', copy=False)
log_EVT_square = pd.read_csv('log_EVT_square.csv', sep='\t')
all_train = pd.merge(all_train, log_EVT_square, on=['USRID'], how='left', copy=False)
all_train.fillna(0, inplace=True)
all_train.to_csv('all_train.csv', sep='\t', index=False)
test_agg = pd.read_csv('test/test_agg.csv', sep='\t')
test_log = pd.read_csv('test/test_log.csv', sep='\t')
EVT_LBL_len, t0, t1, t2, t3, t4, t5, t6, t7, t8, t9 = log_tabel(test_log)
test_set = pd.merge(test_agg, EVT_LBL_len, on=['USRID'], how='left')
test_set = pd.merge(test_set, t0, on=['USRID'], how='left')
test_set = pd.merge(test_set, t1, on=['USRID'], how='left')
test_set = pd.merge(test_set, t2, on=['USRID'], how='left')
test_set = pd.merge(test_set, t3, on=['USRID'], how='left')
test_set = pd.merge(test_set, t4, on=['USRID'], how='left')
test_set = pd.merge(test_set, t5, on=['USRID'], how='left')
test_set = pd.merge(test_set, t6, on=['USRID'], how='left')
test_set = pd.merge(test_set, t7, on=['USRID'], how='left')
test_set = pd.merge(test_set, t8, on=['USRID'], how='left')
test_set = pd.merge(test_set, t9, on=['USRID'], how='left')
test_set = pd.merge(test_set, log_pre, on=['USRID'], how='left', copy=False)
test_set = pd.merge(test_set, log_EVT_square, on=['USRID'], how='left', copy=False)
test_set.fillna(0, inplace=True)
test_set.to_csv('test_set.csv', sep='\t', index=False)
def main():
log_pre()
log_EVT_square()
Merge()
if __name__ == '__main__':
main()