-
Notifications
You must be signed in to change notification settings - Fork 0
/
ModelLoader.py
257 lines (194 loc) · 12.2 KB
/
ModelLoader.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
from Data import *
from calibrators import *
import pickle
from utils import hot_padding
class ModelLoader():
def __init__(self, dataset_name, shuffle_num, model_name, Norm='L2'):
self.dataset_name = dataset_name
self.model_name = model_name
self.shuffle_num = shuffle_num
self.data = load_data(dataset_name, shuffle_num)
calc_dir = f'./{dataset_name}/{shuffle_num}/{model_name}/'
self.all_predictions_val = np.load(calc_dir + f'all_predictions_val.npy', allow_pickle=True)
self.all_predictions_test = np.load(calc_dir + f'all_predictions_test.npy', allow_pickle=True)
# self.all_predictions_train = np.load(calc_dir + f'all_predictions_train{adder}.npy', allow_pickle=True)
self.y_pred_test = np.load(calc_dir + f'y_pred_test.npy', allow_pickle=True)
self.y_pred_val = np.load(calc_dir + f'y_pred_val.npy', allow_pickle=True)
# self.y_pred_train = np.load(calc_dir + f'y_pred_train{adder}.npy', allow_pickle=True)
# adder of norm:
if Norm in ['L1', 'Linf', 'L2']:
Norm = '_' + Norm
else:
raise ValueError(f'{Norm} not supported')
try:
# self.stability_test = np.load(calc_dir + f'/stability_test{adder}.npy', allow_pickle=True)
# self.stability_val = np.load(calc_dir + f'/stability_val{adder}.npy', allow_pickle=True)
self.stability_test = np.load(calc_dir + f'stability_test{Norm}.npy', allow_pickle=True)
self.stability_val = np.load(calc_dir + f'stability_val{Norm}.npy', allow_pickle=True)
except:
print('couldnt load stability calculations')
self.stability_test = None
self.stability_val = None
try:
self.sep_test = np.load(calc_dir + f'/sep_test{Norm}.npy', allow_pickle=True)
self.sep_val = np.load(calc_dir + f'/sep_val{Norm}.npy', allow_pickle=True)
except:
print('couldnt load saperation calculations')
self.sep_test = None
self.sep_val = None
# for pytorch load of logits
try:
self.logits_test = np.load(calc_dir + f'/logits_test.npy', allow_pickle=True)
self.logits_val = np.load(calc_dir + f'/logits_val.npy', allow_pickle=True)
# self.logits_train = np.load(calc_dir + f'/logits_train.npy', allow_pickle=True)
except:
self.logits_test = None
self.logits_val = None
self.logits_train = None
def compute_error_metric(self, method, err_Func, bins=15):
allowed = ['Base', 'StabilityCalibrator', 'SeparationCalibrator', 'HBCalibrator', 'SBCCalibrator', 'BetaCalibrator', 'BBQCalibrator']
if self.model_name in ['RF', 'GB']:
model_dir = f'{self.dataset_name}/{self.shuffle_num}/model/model_{self.dataset_name}_{self.model_name}.sav'
model = pickle.load(open(model_dir, 'rb'))
allowed.extend(['SKlearn_calibrator_platt', 'SKlearn_calibrator_isotonic'])
elif self.model_name == 'pytorch':
allowed.extend(['IsotonicCalibrator', 'PlattCalibrator', 'TSCalibrator', 'EnsembleTSCalibrator'])
if method in allowed:
if method == 'Base':
# Base
probs_calibrated = self.all_predictions_test
pred_y_test_calibrated = self.y_pred_test
elif method == 'SKlearn_calibrator_isotonic':
calibrator = SKlearn_calibrator(self.data, 'isotonic', model).fit()
probs_calibrated = calibrator.calibrated_model.predict_proba(self.data.X_test)
pred_y_test_calibrated = calibrator.calibrated_model.predict(self.data.X_test)
elif method in ['IsotonicCalibrator', 'PlattCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.all_predictions_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = self.y_pred_test
elif method == 'SKlearn_calibrator_platt':
calibrator = SKlearn_calibrator(self.data, 'sigmoid', model).fit()
probs_calibrated = calibrator.calibrated_model.predict_proba(self.data.X_test)
pred_y_test_calibrated = calibrator.calibrated_model.predict(self.data.X_test)
elif method in ['StabilityCalibrator', 'StabilityHistogramBinningCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.stability_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.stability_test)
pred_y_test_calibrated = self.y_pred_test
elif method in ['SeparationCalibrator', 'SeparationHistogramBinningCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.sep_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.sep_test)
pred_y_test_calibrated = self.y_pred_test
elif method == 'HBCalibrator':
calibrator = HBCalibrator()
calibrator.fit(self.all_predictions_val, self.data.y_val + 1)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = self.y_pred_test
elif method in ['SBCCalibrator', 'BetaCalibrator', 'BBQCalibrator']:
calibrator = SBCCalibrator()
calibrator.fit(self.all_predictions_val, self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = np.argmax(probs_calibrated, axis=1)
elif method in ['EnsembleTSCalibrator', 'TSCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.logits_val, self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = np.argmax(probs_calibrated, axis=1)
# elif method in ['BetaCalibrator', 'BBQCalibrator']:
# calibrator = eval(method)()
# calibrator.fit(self.all_predictions_val, self.data.y_val)
#
# probs_calibrated = calibrator.calibrate(self.all_predictions_test)
# pred_y_test_calibrated = self.y_pred_test
else:
raise ValueError(f'{method} Method do not exist')
return err_Func(probs_calibrated, pred_y_test_calibrated, self.data.y_test, bins)
def __repr__(self):
return '(' + self.dataset_name + '-' + self.model_name + '-' + str(self.shuffle_num) + ')'
class Tabular_loader():
def __init__(self, dataset_name, model_name, shuffle_num, all_predictions_val, all_predictions_test, y_pred_test,
y_pred_val, stability_test, stability_val, data, model):
self.dataset_name = dataset_name
self.model_name = model_name
self.shuffle_num = shuffle_num
self.all_predictions_val = all_predictions_val
self.all_predictions_test = all_predictions_test
self.y_pred_test = y_pred_test
self.y_pred_val = y_pred_val
self.stability_test = stability_test
self.stability_val = stability_val
self.data = data
self.model = model
def compute_error_metric(self, method, err_Func, bins=15):
if method in ['Base', 'StabilityCalibrator', 'SeparationCalibrator', 'HBCalibrator', 'SBCCalibrator',
'SKlearn_calibrator_platt', 'SKlearn_calibrator_isotonic', 'IsotonicCalibrator',
'PlattCalibrator', 'StabilityHistogramBinningCalibrator', 'SeparationHistogramBinningCalibrator']:
if method == 'Base':
# Base
probs_calibrated = self.all_predictions_test
pred_y_test_calibrated = self.y_pred_test
elif method == 'SKlearn_calibrator_isotonic':
calibrator = SKlearn_calibrator(self.data, 'isotonic', self.model).fit()
probs_calibrated = calibrator.calibrated_model.predict_proba(self.data.X_test)
pred_y_test_calibrated = calibrator.calibrated_model.predict(self.data.X_test)
elif method in ['IsotonicCalibrator', 'PlattCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.all_predictions_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = self.y_pred_test
elif method == 'SKlearn_calibrator_platt':
calibrator = SKlearn_calibrator(self.data, 'sigmoid', self.model).fit()
probs_calibrated = calibrator.calibrated_model.predict_proba(self.data.X_test)
pred_y_test_calibrated = calibrator.calibrated_model.predict(self.data.X_test)
elif method in ['StabilityCalibrator', 'StabilityHistogramBinningCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.stability_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.stability_test)
pred_y_test_calibrated = self.y_pred_test
elif method in ['SeparationCalibrator', 'SeparationHistogramBinningCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.sep_val, self.y_pred_val == self.data.y_val)
probs_calibrated = calibrator.calibrate(self.sep_test)
pred_y_test_calibrated = self.y_pred_test
elif method == 'HBCalibrator':
calibrator = HBCalibrator()
calibrator.fit(self.all_predictions_val, self.data.y_val + 1)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = self.y_pred_test
elif method == 'SBCCalibrator':
calibrator = SBCCalibrator()
calibrator.fit(self.all_predictions_val, self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = np.argmax(probs_calibrated, axis=1)
elif method in ['EnsembleTSCalibrator', 'TSCalibrator']:
calibrator = eval(method)()
calibrator.fit(self.logits_val, self.data.y_val)
probs_calibrated = calibrator.calibrate(self.all_predictions_test)
pred_y_test_calibrated = np.argmax(probs_calibrated, axis=1)
elif method == 'stab->SBC':
calibrator = StabilityCalibrator()
calibrator.fit(self.stability_val, self.y_pred_val == self.data.y_val)
stab_val_probs = calibrator.calibrate(self.stability_val)
stab_test_probs = calibrator.calibrate(self.stability_test)
stab_val_probs = hot_padding(stab_val_probs, self.y_pred_val, self.data.num_labels)
stab_test_probs = hot_padding(stab_test_probs, self.y_pred_test, self.data.num_labels)
calibrator = SBCCalibrator()
calibrator.fit(stab_val_probs, self.data.y_val)
probs_calibrated = calibrator.calibrate(stab_test_probs)
pred_y_test_calibrated = np.argmax(probs_calibrated, axis=1)
elif method == 'stab->HB':
calibrator = StabilityCalibrator()
calibrator.fit(self.stability_val, self.y_pred_val == self.data.y_val)
stab_val_probs = calibrator.calibrate(self.stability_val)
stab_test_probs = calibrator.calibrate(self.stability_test)
stab_val_probs = hot_padding(stab_val_probs, self.y_pred_val, self.data.num_labels)
stab_test_probs = hot_padding(stab_test_probs, self.y_pred_test, self.data.num_labels)
calibrator = HBCalibrator()
calibrator.fit(stab_val_probs, self.data.y_val + 1)
probs_calibrated = calibrator.calibrate(stab_test_probs)
pred_y_test_calibrated = self.y_pred_test
else:
raise ValueError(f'{method} Method do not exist')
return err_Func(probs_calibrated, pred_y_test_calibrated, self.data.y_test, bins)