-
Notifications
You must be signed in to change notification settings - Fork 0
/
models_utils.py
421 lines (332 loc) · 17.3 KB
/
models_utils.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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
import pandas as pd
import numpy as np
import pickle
from sklearn.preprocessing import OneHotEncoder, FunctionTransformer, StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.utils import parallel_backend
from sklearn.model_selection import ShuffleSplit, StratifiedShuffleSplit, GridSearchCV, RandomizedSearchCV
import XGB, ExplainableBoosting, ExplainableBoostingRegressor, XGBRegressor, \
SKLearnGBT, SKLearnGBTRegressor, SplineLogisticGAM, SplineGAM, \
LogisticRegressionCV, LinearRegressionRidgeCV, MarginalLogisticRegressionCV, MarginalLinearRegressionCV, \
IndicatorLinearRegressionCV, IndicatorLogisticRegressionCV, FLAM, FLAMRegressor, \
RSpline, RSplineRegressor, RandomForest, RandomForestRegressor, \
XGBLabelEncoding, XGBLabelEncodingRegressor, \
Bagging, BaggingRegressor, BaggingLabelEncoding, BaggingLabelEncodingRegressor, \
OnehotExplainableBoostingRegressor, OnehotExplainableBoosting
import OnehotEncodingFitMixin, LabelEncodingFitMixin, EncodingBase
import Timer
def get_rf_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'rf', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = RandomForestRegressor if problem == 'regression' else RandomForest
# Understand the format
params = {'random_state': random_state, 'n_jobs': -1}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('n'):
params['n_estimators'] = int(param_str[1:])
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
params.update(kwargs)
model = the_cls(**params)
return model
def _get_lr_model_template(model_name, problem, random_state=1377, cls_model_cls=LogisticRegressionCV, reg_model_cls=LinearRegressionRidgeCV, **kwargs):
assert problem in ['regression', 'classification']
the_cls = reg_model_cls if problem == 'regression' else cls_model_cls
# Understand the format
params = {} if problem == 'regression' else {'random_state': random_state}
is_bag = False
bag_params = {'random_state': random_state, 'n_jobs': None}
split_model_name = model_name.split('-')
for param_str in split_model_name[1:]:
if param_str.startswith('o'):
bag_params['n_estimators'] = int(param_str[1:])
is_bag = True
elif param_str.startswith('r'):
if problem == 'classification':
params['random_state'] = int(param_str[1:])
bag_params['random_state'] = int(param_str[1:])
elif param_str == 'l1':
if problem == 'classification':
params['penalty'] = 'l1'
params['solver'] = 'saga'
elif param_str.startswith('q'):
params['binning_strategy'] = 'quantile'
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
params.update(kwargs)
model = the_cls(**params)
if is_bag:
if problem == 'regression':
bag_cls = BaggingLabelEncodingRegressor if isinstance(model, LabelEncodingFitMixin) else BaggingRegressor
else:
bag_cls = BaggingLabelEncoding if isinstance(model, LabelEncodingFitMixin) else Bagging
model = bag_cls(base_estimator=model, **bag_params)
return model
def get_lr_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'lr', 'the model_name is not in the supported format %s' % model_name
reg_model_cls = LinearRegressionRidgeCV
if model_name.startswith('lr-l1') and problem == 'regression':
raise NotImplementedError('Somehow I can not override LassoCV. It has error key_error: no intercept...Not sure how to fix it')
return _get_lr_model_template(model_name, problem, random_state, cls_model_cls=LogisticRegressionCV, reg_model_cls=reg_model_cls, **kwargs)
def get_mlr_model(model_name, problem, random_state=1377, **kwargs):
''' Get Marginal Logistic Regression '''
assert model_name.split('-')[0] == 'mlr', 'the model_name is not in the supported format %s' % model_name
return _get_lr_model_template(model_name, problem, random_state, cls_model_cls=MarginalLogisticRegressionCV, reg_model_cls=MarginalLinearRegressionCV, **kwargs)
def get_ilr_model(model_name, problem, random_state=1377, **kwargs):
''' Get Indicator Logistic Regression '''
assert model_name.split('-')[0] == 'ilr', 'the model_name is not in the supported format %s' % model_name
return _get_lr_model_template(model_name, problem, random_state, cls_model_cls=IndicatorLogisticRegressionCV, reg_model_cls=IndicatorLinearRegressionCV, **kwargs)
def get_xgb_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'xgb', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = XGBRegressor if problem == 'regression' else XGB
# Understand the format
params = {'random_state': random_state}
is_bag = False
bag_params = {'random_state': random_state, 'n_jobs': None}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('d'):
params['max_depth'] = int(param_str[1:])
elif param_str == 'l': # label-encoding instead of one-hot
the_cls = XGBLabelEncodingRegressor if problem == 'regression' else XGBLabelEncoding
elif param_str.startswith('o'):
bag_params['n_estimators'] = int(param_str[1:])
is_bag = True
elif param_str.startswith('cols'): # 'cols0.9'
params['colsample_bytree'] = float(param_str[4:])
elif param_str.startswith('cv'):
continue
elif param_str.startswith('reg'):
params['reg_lambda'] = float(param_str[3:])
elif param_str.startswith('r'):
params['random_state'] = int(param_str[1:])
bag_params['random_state'] = int(param_str[1:])
elif param_str.startswith('cw'):
params['min_child_weight'] = float(param_str[2:])
elif param_str.startswith('lr'):
params['learning_rate'] = float(param_str[2:])
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
params.update(kwargs)
model = the_cls(**params)
if is_bag:
if problem == 'regression':
bag_cls = BaggingLabelEncodingRegressor if isinstance(model, LabelEncodingFitMixin) else BaggingRegressor
else:
bag_cls = BaggingLabelEncoding if isinstance(model, LabelEncodingFitMixin) else Bagging
model = bag_cls(base_estimator=model, **bag_params)
return model
def get_skgbt_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'skgbt', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = SKLearnGBTRegressor if problem == 'regression' else SKLearnGBT
# Understand the format
params = {}
bag_params, is_bag = {'random_state': random_state, 'n_jobs': -1}, False
for param_str in model_name.split('-')[1:]:
if param_str.startswith('o'):
bag_params['n_estimators'] = int(param_str[1:])
is_bag = True
elif param_str.startswith('cv'):
continue
elif param_str.startswith('d'):
params['max_depth'] = int(param_str[1:])
elif param_str.startswith('r'):
bag_params['random_state'] = int(param_str[1:])
elif param_str == 'v2':
continue # skgbt-v2
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
params.update(kwargs)
model = the_cls(**params)
if is_bag:
if problem == 'regression':
bag_cls = BaggingLabelEncodingRegressor if isinstance(model, LabelEncodingFitMixin) else BaggingRegressor
else:
bag_cls = BaggingLabelEncoding if isinstance(model, LabelEncodingFitMixin) else Bagging
model = bag_cls(base_estimator=model, **bag_params)
return model
def get_spline_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'spline', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = SplineGAM if problem == 'regression' else SplineLogisticGAM
# Understand the format
spline_params = {'search': True}
is_bag = False
bag_params = {'random_state': random_state, 'n_jobs': None}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('lam'):
spline_params['lam'] = float(param_str[3:])
spline_params['search'] = False
elif param_str == 'b': # spline-b
spline_params['fit_binary_feat_as_factor_term'] = True
elif param_str.startswith('cv'):
continue
elif param_str.startswith('o'):
bag_params['n_estimators'] = int(param_str[1:])
is_bag = True
elif param_str == 'v2':
continue # spline-v2
elif param_str.startswith('r'):
bag_params['random_state'] = int(param_str[1:])
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
spline_params.update(kwargs)
model = the_cls(**spline_params)
if is_bag:
if problem == 'regression':
bag_cls = BaggingLabelEncodingRegressor if isinstance(model, LabelEncodingFitMixin) else BaggingRegressor
else:
bag_cls = BaggingLabelEncoding if isinstance(model, LabelEncodingFitMixin) else Bagging
model = bag_cls(base_estimator=model, **bag_params)
return model
def get_flam_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'flam', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = FLAMRegressor if problem == 'regression' else FLAM
# Understand the format
spline_params = {'random_state': random_state}
is_bag = False
bag_params = {'random_state': random_state, 'n_jobs': None}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('lam'):
spline_params['lam'] = float(param_str[3:])
spline_params['search'] = False
elif param_str.startswith('o'):
bag_params['n_estimators'] = int(param_str[1:])
is_bag = True
elif param_str == 'v':
spline_params['verbose'] = True
elif param_str.startswith('r'):
spline_params['random_state'] = int(param_str[1:])
bag_params['random_state'] = int(param_str[1:])
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
spline_params.update(kwargs)
model = the_cls(**spline_params)
if is_bag:
if problem == 'regression':
bag_cls = BaggingLabelEncodingRegressor if isinstance(model, LabelEncodingFitMixin) else BaggingRegressor
else:
bag_cls = BaggingLabelEncoding if isinstance(model, LabelEncodingFitMixin) else Bagging
model = bag_cls(base_estimator=model, **bag_params)
return model
def get_rspline_model(model_name, problem, random_state=1377, **kwargs):
assert model_name.split('-')[0] == 'rspline', 'the model_name is not in the supported format %s' % model_name
assert problem in ['regression', 'classification']
the_cls = RSplineRegressor if problem == 'regression' else RSpline
# Understand the format
spline_params = {}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('k'):
spline_params['maxk'] = int(param_str[1:])
elif param_str == 's':
spline_params['select'] = True
elif param_str.startswith('nd'): # non-discrete
spline_params['discrete'] = False
elif param_str == 'gam': # rspline-gam
spline_params['discrete'] = False
spline_params['model_to_use'] = 'gam'
elif param_str.startswith('r'):
spline_params['random_state'] = int(param_str[1:])
elif param_str == 'v2':
continue # rspline-v2
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
spline_params.update(kwargs)
model = the_cls(**spline_params)
return model
def get_ebm_model(model_name, problem, random_state=1377, **kwargs):
the_cls = ExplainableBoostingRegressor if problem == 'regression' else ExplainableBoosting
assert model_name.split('-')[0] == 'ebm', 'the model_name is not in the supported format %s' % model_name
# Understand the format
params = {'random_state': random_state}
for param_str in model_name.split('-')[1:]:
if param_str.startswith('o'):
params['n_estimators'] = int(param_str[1:])
elif param_str.startswith('cv'):
continue
elif param_str.startswith('it'):
params['interactions'] = int(param_str[2:])
elif param_str.startswith('i'):
params['feature_step_n_inner_bags'] = int(param_str[1:])
elif param_str.startswith('bf'):
params['feature_fit_scheme'] = 'best_first'
elif param_str.startswith('r'):
params['random_state'] = int(param_str[1:])
elif param_str.startswith('q'):
params['binning_strategy'] = 'quantile'
elif param_str == 'h': # onehot encoding
the_cls = OnehotExplainableBoostingRegressor if problem == 'regression' else OnehotExplainableBoosting
else:
raise NotImplementedError('the param_str is not in the supported format %s' % param_str)
params.update(kwargs)
ebm = the_cls(**params)
return ebm
def get_model(X_train, y_train, problem, model_name, random_state=1377, **kwargs):
assert np.sum([model_name.startswith(k) \
for k in ['ebm', 'spline', 'skgbt', 'xgb', 'lr', 'mlr', 'ilr', 'rf', 'flam', 'rspline']]) == 1, \
'Model name is wierd! %s' % model_name
for k in ['ebm', 'spline', 'skgbt', 'xgb', 'lr', 'mlr', 'ilr', 'rf', 'flam', 'rspline']:
if model_name.startswith(k):
the_model = eval('get_%s_model' % k)(model_name, problem, random_state=random_state, **kwargs)
break
else:
raise RuntimeError('No model class found with name %s' % model_name)
if not hasattr(the_model, 'param_distributions') or the_model.param_distributions is None or '-cv' not in model_name:
the_model.fit(X_train, y_train)
else:
with Timer('Use cv to select hyperparameters'):
cv_cls = StratifiedShuffleSplit if problem == 'classification' else ShuffleSplit
scoring = 'roc_auc' if problem == 'classification' else 'neg_mean_squared_error'
cv = cv_cls(n_splits=3, test_size=0.15, random_state=random_state)
cv_model = RandomizedSearchCV(
the_model, param_distributions=the_model.param_distributions, n_iter=8, n_jobs=8,
scoring=scoring, cv=cv, refit=True, random_state=random_state, error_score=np.nan)
with parallel_backend('loky'):
cv_model.fit(X_train, y_train)
the_model = cv_model.best_estimator_
return the_model
def pickle_load(path, mode='rb'):
''' Since I rename the module from class "models" to "arch", all the pickled file needs special handling '''
with open(path, mode) as fp:
try:
model = pickle.load(fp)
except AttributeError as e:
class CustomUnpickler(pickle.Unpickler):
def find_class(self, module, name):
if module == 'models':
module = 'arch'
return super().find_class(module, name)
model = CustomUnpickler(fp).load()
if not isinstance(model, EncodingBase):
return model
if not hasattr(model, 'cat_columns') and \
('flam' in path or 'rspline' in path or 'spline' in path or 'xgb' in path \
or 'lr' in path):
for d_name in ['adult', 'churn', 'compass', 'heart']:
if d_name in path:
with Timer('Converting old model to new model in path: %s' % path):
model = convert_old_model(model, d_name)
pickle.dump(model, open(path, 'wb'))
break
return model
def convert_old_model(model, d_name):
if hasattr(model, 'cat_columns'):
return model
from loaddata_utils import load_data
X = load_data(d_name)['full']['X']
if isinstance(model, FLAM) or isinstance(model, FLAMRegressor) \
or isinstance(model, RSpline) or isinstance(model, RSplineRegressor): # get the old df, transform and set it back
df = model.get_GAM_plot_dataframe()
model.GAM_plot_dataframe = model.revert_dataframe(df)
if (isinstance(model, BaggingRegressor) or isinstance(model, Bagging)):
model.not_revert = True
model.cat_columns = X.columns[X.dtypes == object].values.tolist()
if isinstance(model, LabelEncodingFitMixin):
raise Exception('Should just discard these models!')
return model
def pickle_dump(model, path):
with open(path, 'wb') as op:
pickle.dump(model, op)