-
Notifications
You must be signed in to change notification settings - Fork 0
/
mlp.py
170 lines (139 loc) · 5.49 KB
/
mlp.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
from datasets import Datasets
from keras.layers import Dense, Dropout
from keras.models import Sequential
from keras.optimizers import SGD, RMSprop
import keras
import time
class TimeHistory(keras.callbacks.Callback):
def on_train_begin(self, logs={}):
self.times = []
def on_epoch_begin(self, batch, logs={}):
self.epoch_time_start = time.time()
def on_epoch_end(self, batch, logs={}):
self.times.append(time.time() - self.epoch_time_start)
class MLP:
def __init__(self, layers, data, optimizer, hid_act='tanh',
out_act='softmax', dropout=None, wdecay=None,
loss='categorical_crossentropy', metrics=['accuracy']):
"""Builds a feedforward neural network. Uses stochastic gradient
descent for optimization.
Parameters:
layers list, number of hidden units in each layer
data data tuple: (x_train, y_train), (x_test, y_test)
Optional:
hid_act hidden activation function
out_act output activation function
dropout dropout value
loss what loss function to use
metrics what metrics to use for evaluation
"""
# check input
assert len(layers) > 0
# preprocess optimizer
if isinstance(optimizer, str):
if optimizer == 'sgd':
optimizer = SGD(lr=0.01, decay=1e-6,
momentum=0.9, nesterov=True)
elif optimizer == 'rms':
optimizer = RMSprop()
else:
raise ValueError('Invalid optimizer')
# preprocess data
if isinstance(data, str):
if data == 'mnist':
data = Datasets.mnist()
# save parameters
self.hid_act = hid_act
self.out_act = out_act
self.loss = loss
self._layers = layers.copy()
# load the data
(self.x_train, self.y_train), (self.x_test, self.y_test) = data
input_dim = self.x_train.shape[1]
output_dim = self.y_train.shape[1]
# preprocess the layers
if dropout:
layers = [int(e / dropout) for e in layers]
# create the model
self.model = Sequential()
# add first hidden layer
layer = Dense(layers[0], activation=hid_act, input_dim=input_dim)
if wdecay:
layer.kernel_regularizer = keras.regularizers.l2(wdecay)
self.model.add(layer)
if dropout:
dropout_layer = Dropout(dropout)
self.model.add(dropout_layer)
# add subsequent layers
for i in range(1, len(layers)):
layer = Dense(layers[i], activation=hid_act)
self.model.add(layer)
if dropout:
dropout_layer = Dropout(dropout)
self.model.add(dropout_layer)
# add output layer
output_layer = Dense(output_dim, activation=out_act)
self.model.add(output_layer)
# setup optimizer
self.optimizer = optimizer
# compile model
self.model.compile(loss=loss,
optimizer=self.optimizer,
metrics=metrics)
def train(self, epochs, batch_size, patience=0):
"""Train the MLP.
Parameters:
epochs how many epochs to train
batch_size size of batches
Returns:
hist history object for training losses
"""
# time callback
callbacks = [
TimeHistory()
]
if patience > 0:
epochs = 1000
callbacks.append(
keras.callbacks.EarlyStopping(monitor='val_accuracy',
patience=patience)
)
# train the model
hist = self.model.fit(self.x_train,
self.y_train,
epochs=epochs,
batch_size=batch_size,
validation_data=(self.x_test, self.y_test),
callbacks=callbacks)
# evaluate model
score = self.model.evaluate(self.x_test,
self.y_test,
batch_size=batch_size)
# log test scores
print('Test loss: {}\nTest acc: {}'.format(
score[0], score[1]))
# extract times
times = callbacks[0].times
# extract history
hist = hist.history
# save all stats in one dictionary
stats = {
'epoch': [], 'time': [], 'val_loss': [],
'val_accuracy': [], 'train_loss': [], 'train_accuracy': [],
'comments': []
}
for i in range(len(times)):
comments = '[{}]'.format(', '.join(map(str, self._layers)))
comments += ', {}, {}, {}, {}, patience {}'.format(
self.hid_act, self.out_act, self.loss, str(self.optimizer),
patience
)
stats['epoch'] += [i + 1]
stats['time'] += ['{:.3f}'.format(times[i])]
stats['val_loss'] += ['{:.4f}'.format(hist['val_loss'][i])]
stats['val_accuracy'] += ['{:.4f}'.format(hist['val_accuracy'][i])]
stats['train_loss'] += ['{:.4f}'.format(hist['loss'][i])]
stats['train_accuracy'] += ['{:.4f}'.format(hist['accuracy'][i])]
stats['comments'] += [comments]
# return statistics dictionary
return stats