forked from alvarouc/mlp
/
model.py
202 lines (173 loc) · 6.75 KB
/
model.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
from keras.models import Sequential
from keras.layers.core import Dense, Dropout
from keras.optimizers import Adadelta
from keras.regularizers import l1
from keras.callbacks import EarlyStopping
from sklearn.cross_validation import StratifiedShuffleSplit
from sklearn.metrics import roc_auc_score, f1_score
from sklearn.base import BaseEstimator, ClassifierMixin
from theano import function
import numpy as np
class BaseMLP(BaseEstimator, ClassifierMixin):
'''
Model class that wraps Keras model
Input:
in_dim: number of variables of the data
out_dim: number of classes to predict
n_hidden: number of hidden variables at each layers
n_deep: number of layers
l1_norm: penalization coefficient for L1 norm on hidden variables
drop: dropout percentage at each layer
verbose: verbosity level (up to 3 levels)
Methods:
reset_weigths: re-initiallizes the weights
save/load: saves/loads weights to a file
fit: trains on data provided with early stopping
train_batch: trains on batch provided splitting
the data into train and validation
fit_batches: trains on a sequence of batches with early stopping
predict: returns prediction on data provided
auc: returns area under the roc curve on data and true
labels provided
'''
def __init__(self, n_hidden=1000, n_deep=4,
l1_norm=0.01, drop=0.1,
early_stop=True, max_epoch=5000,
patience=200,
learning_rate=0.1, verbose=2):
self.max_epoch = max_epoch
self.early_stop = early_stop
self.n_hidden = n_hidden
self.n_deep = n_deep
self.l1_norm = l1_norm
self.drop = drop
self.patience = patience
self.verbose = verbose
self.learning_rate = learning_rate
def fit(self, X, y, **kwargs):
n_class = len(np.unique(y))
if n_class == 2:
out_dim = 1
else:
out_dim = n_class
self.build_model(X.shape[1], out_dim)
#if self.verbose:
temp = [layer['output_dim']
for layer in self.model.get_config()['layers']
if layer['name'] == 'Dense']
print('Model:{}'.format(temp))
print('l1: {}, drop: {}, lr: {}, patience: {}'.format(
self.l1_norm, self.drop, self.learning_rate,
self.patience))
return self
def save(self, path):
self.model.save_weights(path)
def load(self, path):
self.model.load_weights(path)
def build_model(self, in_dim, out_dim):
self.model = build_model(in_dim, out_dim=out_dim,
n_hidden=self.n_hidden, l1_norm=self.l1_norm,
n_deep=self.n_deep, drop=self.drop,
learning_rate=self.learning_rate)
return self
def feed_forward(self, X):
# Feeds the model with X and returns the output of
# each layer
layer_output = []
for layer in self.model.layers:
if layer.get_config()['name'] == 'Dense':
get_layer = function([self.model.layers[0].input],
layer.get_output(train=False),
allow_input_downcast=True)
layer_output.append(get_layer(X))
return layer_output
def predict_proba(self, X):
proba = self.model.predict(X, verbose=self.verbose)
if proba.shape[1] == 1:
proba = np.array(proba).reshape((X.shape[0], -1))
temp = (1-proba.sum(axis=1)).reshape(X.shape[0], -1)
proba = np.hstack((temp, proba))
return proba
def predict(self, X):
prediction = self.model.predict_classes(X, verbose=self.verbose)
prediction = np.array(prediction).reshape((X.shape[0], -1))
prediction = np.squeeze(prediction).astype('int')
return(prediction)
def auc(self, X, y):
prediction = self.predict(X)
return roc_auc_score(y, prediction)
def f1(self, X, y):
n_class = len(np.unique(y))
prediction = self.predict(X)
if n_class > 2:
return f1_score(y, prediction, average='weighted')
else:
return f1_score(y, prediction)
class MLP(BaseMLP):
def fit(self, X, y):
super().fit(X, y)
self.classes_, y = np.unique(y, return_inverse=True)
n_class = len(np.unique(y))
if n_class > 2:
y = np.array([np.roll([1] + [0]*(n_class-1), pos)
for pos in y.astype('int')])
if self.early_stop:
sss = StratifiedShuffleSplit(y, 1, test_size=0.1,
random_state=0)
train_index, val_index = next(iter(sss))
x_train, x_val = X[train_index, :], X[val_index, :]
y_train, y_val = y[train_index], y[val_index]
stop = EarlyStopping(monitor='val_loss',
patience=self.patience,
verbose=self.verbose)
self.model.fit(x_train, y_train,
nb_epoch=self.max_epoch,
# batch_size=64,
verbose=self.verbose,
callbacks=[stop],
show_accuracy=True,
validation_data=(x_val, y_val))
else:
self.model.fit(X, y, nb_epoch=self.max_epoch,
verbose=self.verbose,
show_accuracy=True)
return self
def build_model(in_dim, out_dim=1,
n_hidden=100, l1_norm=0.0,
n_deep=5, drop=0.1,
learning_rate=0.1):
model = Sequential()
# Input layer
model.add(Dense(
input_dim=in_dim,
output_dim=n_hidden,
init='glorot_uniform',
activation='tanh',
W_regularizer=l1(l1_norm)))
# do X layers
for layer in range(n_deep-1):
model.add(Dropout(drop))
model.add(Dense(
output_dim=np.round(n_hidden/2**(layer+1)),
init='glorot_uniform',
activation='tanh',
W_regularizer=l1(l1_norm)))
# Output layer
if out_dim == 1:
activation = 'tanh'
else:
activation = 'softmax'
model.add(Dense(out_dim,
init='glorot_uniform',
activation=activation))
# Optimization algorithms
opt = Adadelta(lr=learning_rate)
if out_dim == 1:
model.compile(loss='binary_crossentropy',
optimizer=opt,
class_mode='binary')
else:
model.compile(loss='categorical_crossentropy',
optimizer=opt,
class_mode='categorical')
return model