-
Notifications
You must be signed in to change notification settings - Fork 0
/
stacked_generalization.py
234 lines (171 loc) · 8.47 KB
/
stacked_generalization.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
import numpy as np
import joblib
import glob
from copy import copy
from sklearn.metrics import f1_score
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import classification_report, accuracy_score, confusion_matrix
from sklearn.base import BaseEstimator, ClassifierMixin
import keras.models as keras_models
from keras.utils.np_utils import to_categorical
from keras.callbacks import ModelCheckpoint, ReduceLROnPlateau
def get_predictions(model, X):
if hasattr(model, 'predict_proba'): # Normal SKLearn classifiers
pred = model.predict_proba(X)
elif hasattr(model, '_predict_proba_lr'): # SVMs
pred = model._predict_proba_lr(X)
else:
pred = model.predict(X)
if len(pred.shape) == 1: # for 1-d ouputs
pred = pred[:, None]
return pred
def check_module_exists(modulename):
try:
__import__(modulename)
except ImportError:
return False
return True
class StackedGeneralizer(BaseEstimator, ClassifierMixin):
"""Base class for stacked generalization classifier models
"""
def __init__(self, blending_models=None, n_folds=10, check_dirs=None, verbose=True):
"""
Stacked Generalizer Classifier
Trains a series of base models using K-fold cross-validation, then combines
the predictions of each model into a set of features that are used to train
a high-level classifier model.
Parameters
-----------
blending_model: object
A classifier model used to aggregate the outputs of the trained base
models. Must have a .fit and .predict_proba/.predict method
n_folds: int
The number of K-folds to use in =cross-validated model training
verbose: boolean
"""
self.blending_models = blending_models
self.n_folds = n_folds
self.check_dirs = check_dirs
self.verbose = verbose
def fit(self, X_indices, y):
X_blend = self._fitTransformBaseModels()
if X_indices is not None:
self._fitBlendingModel(X_blend[X_indices], y)
else:
self._fitBlendingModel(X_blend, y)
def predict(self, pred_directory, X_indices=None):
# perform model averaging to get predictions
predictions_dir = pred_directory if pred_directory is not None else 'models/*/'
X_blend = self.transformBaseModels(predictions_dir)
if X_indices is not None:
predictions = self._transformBlendingModel(X_blend[X_indices])
else:
predictions = self._transformBlendingModel(X_blend)
pred_classes = np.argmax(predictions, axis=1)
return pred_classes
def predict_proba(self, pred_directory, X_indices=None):
# perform model averaging to get predictions
predictions_dir = pred_directory if pred_directory is not None else 'models/*/'
X_blend = self.transformBaseModels(predictions_dir)
if X_indices is not None:
predictions = self._transformBlendingModel(X_blend[X_indices])
else:
predictions = self._transformBlendingModel(X_blend)
return predictions
def transformBaseModels(self, pred_dir='models/*/'):
# predict via model averaging
predictions = []
base_path = pred_dir
path = base_path + "*.npy"
files = glob.glob(path)
for file in files:
if self.check_dirs is not None:
for name in self.check_dirs:
if name[:-1] in file:
if self.verbose: print('Loading numpy file %s' % (file))
cv_predictions = np.load(file)
if 'voting' not in file:
predictions.append(cv_predictions.mean(axis=0)) # take mean on all cv predictions of that model
else:
predictions.append(cv_predictions)
break
continue
else:
if self.verbose: print('Loading numpy file %s' % (file))
cv_predictions = np.load(file)
print('Added file %s' % file)
if 'voting' not in file:
predictions.append(cv_predictions.mean(axis=0)) # take mean on all cv predictions of that model
else:
predictions.append(cv_predictions)
# concat all features
predictions = np.hstack(predictions)
if self.verbose: print('Loaded predictions. Shape : ', predictions.shape)
return predictions
def _fitTransformBaseModels(self):
return self.transformBaseModels()
def _fitBlendingModel(self, X_blend, y):
self.blending_model_cv = []
for model_id, blend_model in enumerate(self.blending_models):
if self.verbose:
model_name = "%s" % blend_model.__repr__()
print('Fitting Blending Model:\n%s' % model_name)
scores = []
skf = StratifiedKFold(self.n_folds, shuffle=True, random_state=1000)
for j, (train_idx, test_idx) in enumerate(skf.split(X_blend, y)):
if self.verbose:
print('Fold %d' % (j + 1))
X_train, y_train = X_blend[train_idx], y[train_idx]
X_test, y_test = X_blend[test_idx], y[test_idx]
if isinstance(blend_model, keras_models.Model) or isinstance(blend_model, keras_models.Sequential):
model = blend_model
model_path = 'stack_model/keras_model_%d_cv_%d' % (model_id + 1, j + 1) + '.h5'
checkpoint = ModelCheckpoint(model_path,
monitor='val_fbeta_score', verbose=1,
save_best_only=True, save_weights_only=True,
mode='max')
reduce_lr = ReduceLROnPlateau(monitor='val_fbeta_score', patience=5, mode='max',
factor=0.8, cooldown=5, min_lr=1e-6, verbose=2)
y_train_categorical = to_categorical(y_train, 3)
y_test_categorical = to_categorical(y_test, 3)
model.fit(X_train, y_train_categorical, batch_size=128, nb_epoch=50,
callbacks=[checkpoint, reduce_lr],
validation_data=(X_test, y_test_categorical))
model.load_weights(model_path)
preds = model.predict(X_test, batch_size=128)
preds = np.argmax(preds, axis=1)
score = f1_score(y_test, preds, average='micro')
scores.append(score)
print('Keras Model %d - CV %d Score : %0.3f' % (model_id + 1, j + 1, score))
else:
model = copy(blend_model)
model_path = 'stack_model/sklearn_model_%d_cv_%d' % (model_id + 1, j + 1) + '.pkl'
model.fit(X_train, y_train)
preds = get_predictions(model, X_test)
preds = np.argmax(preds, axis=1)
score = f1_score(y_test, preds, average='micro')
scores.append(score)
print('SKLearn Model %d - CV %d Score : %0.3f' % (model_id + 1, j + 1, score))
joblib.dump(model, model_path)
# add trained model to list of CV'd models
self.blending_model_cv.append(model)
print('Average F1 score of model : ', sum(scores) / len(scores))
def _transformBlendingModel(self, X_blend):
# make predictions from averaged models
cv_predictions = None
n_models = len(self.blending_model_cv)
for i, model in enumerate(self.blending_model_cv):
if self.verbose: print('Getting predictions from blending model %s (Classifier id %d)' %
(model.__class__.__name__, i + 1))
cv_predictions = None
model_predictions = get_predictions(model, X_blend)
if cv_predictions is None:
cv_predictions = np.zeros((n_models, X_blend.shape[0], model_predictions.shape[1]))
cv_predictions[i, :, :] = model_predictions
# perform model averaging to get predictions
predictions = cv_predictions.mean(0)
return predictions
if __name__ == '__main__':
model = StackedGeneralizer()
X_blend = model.transformBaseModels()
print(X_blend.shape)