-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathgroupby.py
253 lines (229 loc) · 13.9 KB
/
groupby.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
#https://zhuanlan.zhihu.com/p/36583002
#先跑一轮，把特征固化到硬盘内。第二轮只是读取硬盘里的特征数据，做特征拼接来减少内存的消耗。
#这个特征可以引入DataSupport里面最为一个卖点
GROUPBY_AGGREGATIONS = [
{'groupby': ["ip"], 'select': 'app', 'agg': 'nunique'},
{'groupby': ["ip"], 'select': 'app', 'agg': 'count'},
]
def get_global_feature(feature):
print ("get global feat")
data = pd.concat([get_basic_data(day=8),get_basic_data(day=9),get_basic_data(day=10)])
for spec in tqdm(GROUPBY_AGGREGATIONS):
new_feature = '{}_{}_{}'.format('_'.join(spec['groupby']), spec['agg'], spec['select'])
result_path = cache_path + new_feature + '%s.hdf'%(data.shape[0])
if os.path.exists(result_path):
result = pd.read_hdf(result_path, 'w')
result[new_feature] = result[new_feature].astype("float32")
else:
print("Grouping by {}, and aggregating {} with {}".format(spec['groupby'], spec['select'], agg_name))
all_features = list(set(spec['groupby'] + [spec['select']]))
result = data[all_features].groupby(spec['groupby'],as_index=False)[spec['select']].agg({new_feature:spec['agg']})
result[new_feature] = result[new_feature].astype("float32")
result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
feature = feature.merge(result, on=spec['groupby'], how='left'，copy=False)
return feature
# Define all the groupby transformations
GROUPBY_AGGREGATIONS = [
# V1 - GroupBy Features #
#########################
# Variance in day, for ip-app-channel
{'groupby': ['ip','app','channel'], 'select': 'day', 'agg': 'var'},
# Variance in hour, for ip-app-os
{'groupby': ['ip','app','os'], 'select': 'hour', 'agg': 'var'},
# Variance in hour, for ip-day-channel
{'groupby': ['ip','day','channel'], 'select': 'hour', 'agg': 'var'},
# Count, for ip-day-hour
{'groupby': ['ip','day','hour'], 'select': 'channel', 'agg': 'count'},
# Count, for ip-app
{'groupby': ['ip', 'app'], 'select': 'channel', 'agg': 'count'},
# Count, for ip-app-os
{'groupby': ['ip', 'app', 'os'], 'select': 'channel', 'agg': 'count'},
# Count, for ip-app-day-hour
{'groupby': ['ip','app','day','hour'], 'select': 'channel', 'agg': 'count'},
# Mean hour, for ip-app-channel
{'groupby': ['ip','app','channel'], 'select': 'hour', 'agg': 'mean'},
# V2 - GroupBy Features #
#########################
# Average clicks on app by distinct users; is it an app they return to?
{'groupby': ['app'],
'select': 'ip',
'agg': lambda x: float(len(x)) / len(x.unique()),
'agg_name': 'AvgViewPerDistinct'
},
# How popular is the app or channel?
{'groupby': ['app'], 'select': 'channel', 'agg': 'count'},
{'groupby': ['channel'], 'select': 'app', 'agg': 'count'},
# V3 - GroupBy Features #
# https://www.kaggle.com/bk0000/non-blending-lightgbm-model-lb-0-977 #
######################################################################
{'groupby': ['ip'], 'select': 'channel', 'agg': 'nunique'},
{'groupby': ['ip'], 'select': 'app', 'agg': 'nunique'},
{'groupby': ['ip','day'], 'select': 'hour', 'agg': 'nunique'},
{'groupby': ['ip','app'], 'select': 'os', 'agg': 'nunique'},
{'groupby': ['ip'], 'select': 'device', 'agg': 'nunique'},
{'groupby': ['app'], 'select': 'channel', 'agg': 'nunique'},
{'groupby': ['ip', 'device', 'os'], 'select': 'app', 'agg': 'nunique'},
{'groupby': ['ip','device','os'], 'select': 'app', 'agg': 'cumcount'},
{'groupby': ['ip'], 'select': 'app', 'agg': 'cumcount'},
{'groupby': ['ip'], 'select': 'os', 'agg': 'cumcount'},
{'groupby': ['ip','day','channel'], 'select': 'hour', 'agg': 'var'}
]
# Apply all the groupby transformations
for spec in GROUPBY_AGGREGATIONS:
# Name of the aggregation we're applying
agg_name = spec['agg_name'] if 'agg_name' in spec else spec['agg']
# Name of new feature
new_feature = '{}_{}_{}'.format('_'.join(spec['groupby']), agg_name, spec['select'])
# Info
print("Grouping by {}, and aggregating {} with {}".format(
spec['groupby'], spec['select'], agg_name
))
# Unique list of features to select
all_features = list(set(spec['groupby'] + [spec['select']]))
# Perform the groupby
gp = X_train[all_features]. \
groupby(spec['groupby'])[spec['select']]. \
agg(spec['agg']). \
reset_index(). \
rename(index=str, columns={spec['select']: new_feature})
# Merge back to X_total
if 'cumcount' == spec['agg']:
X_train[new_feature] = gp[0].values
else:
X_train = X_train.merge(gp, on=spec['groupby'], how='left')
# Clear memory
del gp
gc.collect()
X_train.head()
#############################################################################
NEW_AGGREGATION_RECIPIES = [
(["CODE_GENDER",
"NAME_EDUCATION_TYPE"], [("AMT_ANNUITY", "max"),
("AMT_CREDIT", "max"),
("EXT_SOURCE_1", "median"),
("EXT_SOURCE_2", "median"),
("OWN_CAR_AGE", "max"),
("OWN_CAR_AGE", "sum"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["CODE_GENDER",
"ORGANIZATION_TYPE"], [("AMT_ANNUITY", "median"),
("AMT_INCOME_TOTAL", "median"),
("DAYS_REGISTRATION", "median"),
("EXT_SOURCE_1", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["CODE_GENDER",
"REG_CITY_NOT_WORK_CITY"], [("AMT_ANNUITY", "median"),
("CNT_CHILDREN", "median"),
("DAYS_ID_PUBLISH", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["CODE_GENDER",
"NAME_EDUCATION_TYPE",
"OCCUPATION_TYPE",
"REG_CITY_NOT_WORK_CITY"], [("EXT_SOURCE_1", "median"),
("EXT_SOURCE_2", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["NAME_EDUCATION_TYPE",
"OCCUPATION_TYPE"], [("AMT_CREDIT", "median"),
("AMT_REQ_CREDIT_BUREAU_YEAR", "median"),
("APARTMENTS_AVG", "median"),
("BASEMENTAREA_AVG", "median"),
("EXT_SOURCE_1", "median"),
("EXT_SOURCE_2", "median"),
("EXT_SOURCE_3", "median"),
("NONLIVINGAREA_AVG", "median"),
("OWN_CAR_AGE", "median"),
("YEARS_BUILD_AVG", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["NAME_EDUCATION_TYPE",
"OCCUPATION_TYPE",
"REG_CITY_NOT_WORK_CITY"], [("ELEVATORS_AVG", "median"),
("EXT_SOURCE_1", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
(["OCCUPATION_TYPE"], [("AMT_ANNUITY", "median"),
("CNT_CHILDREN", "median"),
("CNT_FAM_MEMBERS", "median"),
("DAYS_BIRTH", "median"),
("DAYS_EMPLOYED", "median"),
("DAYS_ID_PUBLISH", "median"),
("DAYS_REGISTRATION", "median"),
("EXT_SOURCE_1", "median"),
("EXT_SOURCE_2", "median"),
("EXT_SOURCE_3", "median"),
("NEW_CREDIT_TO_ANNUITY_RATIO", "median"),
("NEW_SOURCES_MEAN", "median"),
("NEW_CREDIT_TO_GOODS_RATIO", "median"),
("NEW_SOURCES_PROD", "median"),
("NEW_CAR_TO_EMPLOY_RATIO", "median"),
("NEW_PHONE_TO_BIRTH_RATIO", "median"),
("NEW_SOURCES_STD", "median"),
("NEW_ANNUITY_TO_INCOME_RATIO", "median"),
("NEW_EMPLOY_TO_BIRTH_RATIO", "median"),
("NEW_PHONE_TO_EMPLOY_RATIO", "median")]),
]
for groupby_cols, specs in NEW_AGGREGATION_RECIPIES:
group_object = self.__application_train.groupby(groupby_cols)
for select, agg in specs:
groupby_aggregate_name = "{}_{}_{}_{}".format("NEW", "_".join(groupby_cols), agg, select)
self.__application_train = self.__application_train.merge(
group_object[select]
.agg(agg)
.reset_index()
.rename(index=str, columns={select: groupby_aggregate_name}),
left_on=groupby_cols,
right_on=groupby_cols,
how="left"
)