-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathfeature_selector.py
322 lines (274 loc) · 13.9 KB
/
feature_selector.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
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
'''
Copyright (C) 2019 ts302_team
Jian Sun: [email protected]
Chunmeng Zhong: [email protected]
Hao Zhang: [email protected]
Hongyu Jia: [email protected]
Xiao Huang: [email protected]
Bin Lin: [email protected]
Zaiyu Pang: [email protected]
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
'''
import lightgbm as lgb
import copy
import hyperopt
from hyperopt import STATUS_OK, Trials, hp, space_eval, tpe
from typing import Dict
from util import *
import CONSTANT
class FeatureSelector:
"""
参数：
data : dataframe 训练数据
labels : array or series, 标签数据，可为None
"""
def __init__(self, data, labels=None):
self.labels = labels
# self.data = copy.deepcopy(data)
self.data = data
self.base_features = list(data.columns)
print(f'FeatureSelector base features:{len(self.base_features)}')
self.feature_importances = None
self.ops = {}
@timeit
def identify_collinear(self, correlation_threshold):
"""基于皮尔逊相关系数识别共线特征"""
num_col = [c for c in self.data if c.startswith(CONSTANT.NUMERICAL_PREFIX) or c.startswith(CONSTANT.CAT_INT_PREFIX)] #只对数值类型做共线特征
if len(num_col) == 0:
self.ops['collinear'] = []
return
if len(num_col) > 1000 and len(self.data) > 20000:
data = data_sample(self.data, 8000)
elif len(self.data) >= 50000:
data = data_sample(self.data, 10000)
else:
data = self.data
corr_matrix = data[num_col].corr()
# 获取相关性矩阵上三角
upper = corr_matrix.where(np.triu(np.ones(corr_matrix.shape), k=1).astype(np.bool))
to_drop = [column for column in upper.columns if any(upper[column].abs() >= correlation_threshold)]
self.ops['collinear'] = to_drop
print('%d features with a correlation magnitude greater than %0.3f.' % (len(self.ops['collinear']), correlation_threshold))
@timeit
def hyperopt_lightgbm(self, X_train: pd.DataFrame, y_train: pd.Series, X_val: pd.DataFrame, y_val: pd.Series,
params: Dict):
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
space = {
"learning_rate": hp.loguniform("learning_rate", np.log(0.01), np.log(0.2)),
"scale_pos_weight": hp.uniform('x', 0, 5),
}
def objective(hyperparams):
model = lgb.train({**params, **hyperparams}, train_data, 500,
valid_data, early_stopping_rounds=100, verbose_eval=0)
score = model.best_score["valid_0"][params["metric"]]
feature_importance_df = pd.DataFrame()
feature_importance_df["features"] = X_train.columns
feature_importance_df["importance_gain"] = model.feature_importance(importance_type='gain')
record_zero_importance = feature_importance_df[feature_importance_df["importance_gain"] == 0.0]
to_drop = list(record_zero_importance['features'])
# in classification, less is better
return {'loss': -score, 'status': STATUS_OK, "drop_feature": to_drop, "best_iter": model.best_iteration}
trials = Trials()
best = hyperopt.fmin(fn=objective, space=space, trials=trials,
algo=tpe.suggest, max_evals=10, verbose=1,
rstate=np.random.RandomState(1))
hyperparams = space_eval(space, best)
log(f"hyperopt auc = {-trials.best_trial['result']['loss']:0.4f} {hyperparams}")
drop_feature = set(X_train.columns.tolist())
for result in trials.results:
drop_feature = drop_feature & set(result['drop_feature'])
return drop_feature, hyperparams
@timeit
def identify_zero_importance(self, valid_ratio=0.2, not_need_list = []):
"""
识别模型训练中零重要性特征
目前使用automl中训练方式，用少量数据集和迭代次数识别零重要性特征
"""
num_feats = [col for col in self.data.columns if
not (col.startswith(CONSTANT.CATEGORY_PREFIX) or col.startswith(
CONSTANT.MULTI_CAT_PREFIX) or col.startswith(CONSTANT.TIME_PREFIX)) and col not in not_need_list]
if len(num_feats) == 0:
self.ops['zero_importance'] = []
return None
X_train, X_val, y_train, y_val = data_split(self.data[num_feats], self.labels, valid_ratio)
print(f'x_train_len: {len(X_train)} x_val_len: {len(X_val)}')
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4,
"max_depth": 6,
"num_leaves": 32,
"feature_fraction": 0.6,
"bagging_fraction": 0.8,
"bagging_freq": 5,
"reg_alpha": 0.1,
"reg_lambda": 0.1,
}
drop_feature, hyperparams = self.hyperopt_lightgbm(X_train, y_train, X_val, y_val, params)
#for c in drop_feature:
# print(c)
self.ops['zero_importance'] = list(drop_feature)
print('%d features with zero importance.\n' % len(self.ops['zero_importance']))
return {**params, **hyperparams}
@timeit
def identify_low_importance(self, top_k=None, top_ratio=None, free_list=[], valid_ratio=0.2, num_boost_round=500, params=None, not_need_list=[]):
log(f"free_list length : {len(free_list)}")
log(f"top_k : {top_k}")
"""
根据比例 or 个数选取重要性靠前特征（识别出重要性靠后特征）
top_k: 个数选取
top_ratio: 比例选取
均不为None则选取两者中较少特征
"""
num_feats = [col for col in self.data.columns if
not (col.startswith(CONSTANT.CATEGORY_PREFIX) or col.startswith(CONSTANT.MULTI_CAT_PREFIX) \
or col.startswith(CONSTANT.TIME_PREFIX) or col.startswith(CONSTANT.CATEGORY_HIGH_PREFIX)) and col not in not_need_list]
if len(num_feats) == 0:
self.ops['low_importance'] = []
self.ops['needed_cols'] = free_list
return
X_train, X_val, y_train, y_val = data_split(self.data[num_feats], self.labels, valid_ratio)
if params == None:
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4,
"max_depth": 6,
"num_leaves": 32,
"feature_fraction": 0.6,
"bagging_fraction": 0.8,
"bagging_freq": 5,
"reg_alpha": 0.1,
"reg_lambda": 0.1,
"learning_rate": 0.01,
#"min_child_weight": 5,
}
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
model = lgb.train(params,
train_data,
num_boost_round,
valid_data,
verbose_eval=0)
self.feature_importances = pd.DataFrame()
self.feature_importances['features'] = X_train.columns
self.feature_importances['importance_gain'] = model.feature_importance(importance_type='gain')
self.feature_importances.sort_values('importance_gain', inplace=True, ascending=False)
total_features_without_zero = len(self.feature_importances[self.feature_importances['importance_gain'] > 0])
needed_cols = free_list.copy()
for i in range(total_features_without_zero):
if len(needed_cols) >= top_k:
break
if self.feature_importances.iloc[i]['importance_gain'] < CONSTANT.MIN_FEAT_IMPORTANT:
break
cur_col = self.feature_importances.iloc[i]['features']
if cur_col in needed_cols:
continue
needed_cols.append(cur_col)
self.ops['low_importance'] = list(set(list(X_train.columns)) - set(needed_cols))
self.ops['needed_cols']=needed_cols
#for c in self.ops['low_importance']:
# print(c)
log('%d features with low importance.\n' % len(self.ops['low_importance']))
@timeit
def hyperopt_cat_lightgbm(self, X_train: pd.DataFrame, y_train: pd.Series, X_val: pd.DataFrame, y_val: pd.Series,
params: Dict, topk=6, free_cat_list=[]):
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
space = {
"learning_rate": hp.loguniform("learning_rate", np.log(0.01), np.log(0.2)),
"scale_pos_weight": hp.uniform('x', 0, 5),
"max_cat_group ": hp.choice("max_cat_group", np.linspace(0, 2000, 20, dtype=int)),
"max_cat_threshold": hp.choice("max_cat_threshold", np.linspace(0, 2000, 20, dtype=int)),
"cat_smooth": hp.choice("cat_smooth", np.linspace(0, 10000, 20, dtype=int)),
}
def objective(hyperparams):
model = lgb.train({**params, **hyperparams}, train_data, 500,
valid_data, early_stopping_rounds=100, verbose_eval=0)
score = model.best_score["valid_0"][params["metric"]]
feature_importance_df = pd.DataFrame()
feature_importance_df["features"] = X_train.columns
feature_importance_df["importance_gain"] = model.feature_importance(importance_type='gain')
# in classification, less is better
return {'loss': -score, 'status': STATUS_OK, "sort_feats": feature_importance_df, "best_iter": model.best_iteration}
trials = Trials()
best = hyperopt.fmin(fn=objective, space=space, trials=trials,
algo=tpe.suggest, max_evals=10, verbose=1,
rstate=np.random.RandomState(1))
hyperparams = space_eval(space, best)
log(f"hyperopt auc = {-trials.best_trial['result']['loss']:0.4f} {hyperparams}")
new_cat_feats = pd.DataFrame([])
for result in trials.results:
feature_import = result['sort_feats'].set_index('features')
feature_import.drop(free_cat_list, axis=0, inplace=True, errors='ignore')
if len(new_cat_feats) == 0:
new_cat_feats = feature_import
else:
new_cat_feats = new_cat_feats + feature_import
new_cat_feats = new_cat_feats[new_cat_feats['importance_gain'] > CONSTANT.MIN_CAT_FEAT_IMPORTANT * max(len(trials.results)/ 2, 1)]
new_cat_feats.sort_values('importance_gain', inplace=True, ascending=False)
if len(new_cat_feats) < topk:
log(f"Test: test pass")
return new_cat_feats.index.tolist(), hyperparams
return new_cat_feats.index.tolist()[0:topk], hyperparams
@timeit
def identify_low_cat_importance(self, valid_ratio=0.2, topk=6, free_cat_list=[]):
cat_feats = [col for col in self.data.columns if col.startswith(CONSTANT.CATEGORY_PREFIX) or col.startswith(CONSTANT.CATEGORY_HIGH_PREFIX)]
if len(cat_feats) == 0:
return []
# self.data[cat_feats] = self.data[cat_feats].astype('category')
# X_train, X_val, y_train, y_val = data_split(self.data[cat_feats], self.labels, valid_ratio)
tmp_data = self.data[cat_feats].astype('category')
X_train, X_val, y_train, y_val = data_split(tmp_data[cat_feats], self.labels, valid_ratio)
print(f'x_train_len: {len(X_train)} x_val_len: {len(X_val)}')
if len(CONSTANT.cat_params) == 0:
params = {
"objective": "binary",
"metric": "auc",
"verbosity": -1,
"seed": 1,
"num_threads": 4,
"max_depth": 6,
"num_leaves": 32,
"feature_fraction": 0.6,
"bagging_fraction": 0.8,
"bagging_freq": 5,
"reg_alpha": 0.1,
"reg_lambda": 0.1,
}
save_cat_features, hyperparams = self.hyperopt_cat_lightgbm(X_train, y_train, X_val, y_val, params, topk, free_cat_list)
CONSTANT.cat_params = {**params, **hyperparams}
return save_cat_features
else:
train_data = lgb.Dataset(X_train, label=y_train)
valid_data = lgb.Dataset(X_val, label=y_val)
model = lgb.train(CONSTANT.cat_params,
train_data,
500,
valid_data,
early_stopping_rounds=100,
verbose_eval=100)
feature_import = pd.DataFrame()
feature_import["features"] = X_train.columns
feature_import["importance_gain"] = model.feature_importance(importance_type='gain')
feature_import.set_index('features', inplace=True)
feature_import.drop(free_cat_list, axis=0, inplace=True, errors='ignore')
feature_import = feature_import[feature_import['importance_gain'] > CONSTANT.MIN_CAT_FEAT_IMPORTANT]
feature_import.sort_values('importance_gain', inplace=True, ascending=False)
if len(feature_import) < topk:
return feature_import.index.tolist()
return feature_import.index.tolist()[0:topk]