This repository has been archived by the owner on Feb 8, 2021. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 7
/
cifar10_keras_mlp3_wd.py
114 lines (102 loc) · 4.57 KB
/
cifar10_keras_mlp3_wd.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
#!/bin/python2
import numpy as np
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense, Flatten
from keras.callbacks import TensorBoard, EarlyStopping
from keras.optimizers import SGD
from keras.initializers import Constant
from keras.regularizers import l2
import keras.backend as K
import pickle
import time
from copy import deepcopy
from shutil import copy
model = Sequential()
model.add(Flatten(input_shape=(28, 28, 3)))
model.add(Dense(512, kernel_initializer='glorot_normal',
bias_initializer=Constant(0.1), kernel_regularizer=l2(1e-4),
activation='relu'))
model.add(Dense(512, kernel_initializer='glorot_normal',
bias_initializer=Constant(0.1), kernel_regularizer=l2(1e-4),
activation='relu'))
model.add(Dense(512, kernel_initializer='glorot_normal',
bias_initializer=Constant(0.1), kernel_regularizer=l2(1e-4),
activation='relu'))
model.add(Dense(10, kernel_initializer='glorot_normal',
bias_initializer=Constant(0.1), activation='softmax'))
early_stop = EarlyStopping(monitor='loss', min_delta=0.0001, patience=5)
now = str(time.time())
tb_callback = TensorBoard(log_dir='../Tensorboard/mlp3/' + now)
img = tf.placeholder(tf.float32, [28, 28, 3])
norm_image = tf.image.per_image_standardization(img)
opt = SGD(lr=0.01, momentum=0.9)
model.compile(loss='mse', optimizer=opt, metrics=['accuracy'])
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
cifar10_train_images = []
cifar10_train_labels = []
print "Loading training images..."
for i in range(1, 6):
train_file = open('../../cifar-10-batches-py/data_batch_' + str(i), 'r')
train_dict = pickle.load(train_file)
for image, label in zip(train_dict['data'], train_dict['labels']):
image_red = np.reshape(image[:1024], (32, 32))[2:-2, 2:-2] / 255.0
image_red = np.reshape(image_red, (28, 28, 1))
image_green = np.reshape(image[1024:2048], (32, 32))[2:-2,
2:-2] / 255.0
image_green = np.reshape(image_green, (28, 28, 1))
image_blue = np.reshape(image[2048:3072], (32, 32))[2:-2, 2:-2] / 255.0
image_blue = np.reshape(image_blue, (28, 28, 1))
image = np.concatenate([image_red, image_green, image_blue], axis=-1)
image = norm_image.eval(feed_dict={img:image})
cifar10_train_images.append(image)
label = np.identity(10)[label]
cifar10_train_labels.append(label)
train_file.close()
epochs = 100
batch_size = 16
prev_loss = 1e4
patience = deepcopy(early_stop.patience)
for epoch in range(epochs):
hist = model.fit(np.array(cifar10_train_images), np.array(
cifar10_train_labels), epochs=(epoch + 1),
batch_size=batch_size, initial_epoch=epoch,
callbacks=[tb_callback])
K.set_value(opt.lr, 0.95 * K.get_value(opt.lr))
if hist.history[early_stop.monitor][0] - prev_loss > early_stop.min_delta:
patience -= 1
else:
patience = deepcopy(early_stop.patience)
if patience <= 0:
break
else:
prev_loss = hist.history[early_stop.monitor][0]
del cifar10_train_images, cifar10_train_labels
print "Loading test images..."
cifar10_test_images = []
cifar10_test_labels = []
test_file = open('../../cifar-10-batches-py/test_batch', 'r')
test_dict = pickle.load(test_file)
for image, label in zip(test_dict['data'], test_dict['labels']):
image_red = np.reshape(image[:1024], (32, 32))[2:-2, 2:-2] / 255.0
image_red = np.reshape(image_red, (28, 28, 1))
image_green = np.reshape(image[1024:2048], (32, 32))[2:-2,
2:-2] / 255.0
image_green = np.reshape(image_green, (28, 28, 1))
image_blue = np.reshape(image[2048:3072], (32, 32))[2:-2, 2:-2] / 255.0
image_blue = np.reshape(image_blue, (28, 28, 1))
image_blue = np.reshape(image_blue, (28, 28, 1))
image = np.concatenate([image_red, image_green, image_blue], axis=-1)
image = norm_image.eval(feed_dict={img:image})
cifar10_test_images.append(image)
label = np.identity(10)[label]
cifar10_test_labels.append(label)
test_file.close()
print(model.evaluate(np.array(cifar10_test_images),
np.array(cifar10_test_labels), batch_size=256))
response = raw_input("Do you want to save this model? (Y/n): ")
if response.lower() not in ['n', 'no', 'nah', 'nein', 'nahi', 'nope']:
model.save('cifar10_mlp3_wd.h5')
copy('./cifar10_keras_mlp3_wd.py', '../Tensorboard/mlp3/' + now)
print "Model saved"