*Neural Nets(NN) for MNIST :


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
from tensorflow.examples.tutorials.mnist import input_data
 
import tensorflow as tf
import random
import matplotlib.pylab as plt
 
learning_rate = 0.001
training_epochs = 15
batch_size = 100
display_step = 1
 
mnist = input_data.read_data_sets('./MNIST_data/', one_hot=True)
 
sess = tf.InteractiveSession()
 
# Create the model
X = tf.placeholder(tf.float32, [None, 784]) #열만 784개로 맞춰라
Y = tf.placeholder(tf.float32, [None, 10])  #열만 10개로 맞춰라
 
W1 = tf.Variable(tf.random_normal([784, 256]))
W2 = tf.Variable(tf.random_normal([256, 256]))
W3 = tf.Variable(tf.random_normal([256, 10]))
 
b1 = tf.Variable(tf.random_normal([256]))
b2 = tf.Variable(tf.random_normal([256]))
b3 = tf.Variable(tf.random_normal([10]))
 
L1 = tf.nn.relu(tf.add(tf.matmul(X, W1), b1))
L2 = tf.nn.relu(tf.add(tf.matmul(L1, W2), b2))
hypothesis = tf.add(tf.matmul(L2, W3), b3)
 
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=hypothesis, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
 
init = tf.initialize_all_variables()
 
with tf.Session() as sess:
    sess.run(init)
 
    for epoch in range (training_epochs):
        avg_cost = 0.
        total_batch = int(mnist.train.num_examples/batch_size)
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            sess.run(optimizer, feed_dict={X:batch_xs, Y:batch_ys})
            avg_cost += sess.run(cost, feed_dict={X: batch_xs, Y:batch_ys})/total_batch
        if epoch % display_step ==0:
            print("Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(avg_cost))
 
    print("Optimization Finished")
 
    correct_prediction = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
 
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
    print("Accuracy:", accuracy.eval({X: mnist.test.images, Y: mnist.test.labels}))
cs


*Xavier initialization :


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
from tensorflow.examples.tutorials.mnist import input_data
 
import tensorflow as tf
import random
import matplotlib.pylab as plt
 
learning_rate = 0.001
training_epochs = 15
batch_size = 100
display_step = 1
 
mnist = input_data.read_data_sets('./MNIST_data/', one_hot=True)
 
sess = tf.InteractiveSession()
 
 
def xavier_init(n_inputs, n_outputs, uniform=True):
    if uniform:
         init_range = tf.sqrt(6.0 / (n_inputs + n_outputs))
         return tf.random_uniform_initializer(-init_range, init_range)
    else:
         stddev = tf.sqrt(3.0 / (n_inputs + n_outputs))
         return tf.truncated_normal_initializer(stddev=stddev)
 
# Create the model
X = tf.placeholder(tf.float32, [None, 784]) #열만 784개로 맞춰라
Y = tf.placeholder(tf.float32, [None, 10])  #열만 10개로 맞춰라
 
# W1 = tf.Variable(tf.random_normal([784, 256]))
# W2 = tf.Variable(tf.random_normal([256, 256]))
# W3 = tf.Variable(tf.random_normal([256, 10]))
W1 = tf.get_variable("W1", shape=[784, 256], initializer=xavier_init(784, 256))
W2 = tf.get_variable("W2", shape=[256, 256], initializer=xavier_init(784, 256))
W3 = tf.get_variable("W3", shape=[256, 10], initializer=xavier_init(784, 256))
 
 
b1 = tf.Variable(tf.random_normal([256]))
b2 = tf.Variable(tf.random_normal([256]))
b3 = tf.Variable(tf.random_normal([10]))
 
L1 = tf.nn.relu(tf.add(tf.matmul(X, W1), b1))
L2 = tf.nn.relu(tf.add(tf.matmul(L1, W2), b2))
hypothesis = tf.add(tf.matmul(L2, W3), b3)
 
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=hypothesis, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
 
 
 
init = tf.initialize_all_variables()
 
with tf.Session() as sess:
    sess.run(init)
 
    for epoch in range (training_epochs):
        avg_cost = 0.
        total_batch = int(mnist.train.num_examples/batch_size)
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            sess.run(optimizer, feed_dict={X:batch_xs, Y:batch_ys})
            avg_cost += sess.run(cost, feed_dict={X: batch_xs, Y:batch_ys})/total_batch
        if epoch % display_step ==0:
            print("Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(avg_cost))
 
    print("Optimization Finished")
 
    correct_prediction = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
 
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
    print("Accuracy:", accuracy.eval({X: mnist.test.images, Y: mnist.test.labels}))
cs





* Mnist 5NN :


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
import tensorflow as tf
import random
import matplotlib.pyplot as plt
 
from tensorflow.examples.tutorials.mnist import input_data
 
tf.set_random_seed(777)  # reproducibility
 
mnist = input_data.read_data_sets("./MNIST_data/", one_hot=True)
 
# parameters
learning_rate = 0.001
training_epochs = 15
batch_size = 100
 
# input place holders
X = tf.placeholder(tf.float32, [None, 784])
Y = tf.placeholder(tf.float32, [None, 10])
 
# dropout (keep_prob) rate  0.7 on training, but should be 1 for testing
keep_prob = tf.placeholder(tf.float32)
 
# weights & bias for nn layers
# http://stackoverflow.com/questions/33640581/how-to-do-xavier-initialization-on-tensorflow
W1 = tf.get_variable("W1", shape=[784, 512],
                     initializer=tf.contrib.layers.xavier_initializer())
b1 = tf.Variable(tf.random_normal([512]))
L1 = tf.nn.relu(tf.matmul(X, W1) + b1)
L1 = tf.nn.dropout(L1, keep_prob=keep_prob)
 
W2 = tf.get_variable("W2", shape=[512, 512],
                     initializer=tf.contrib.layers.xavier_initializer())
b2 = tf.Variable(tf.random_normal([512]))
L2 = tf.nn.relu(tf.matmul(L1, W2) + b2)
L2 = tf.nn.dropout(L2, keep_prob=keep_prob)
 
W3 = tf.get_variable("W3", shape=[512, 512],
                     initializer=tf.contrib.layers.xavier_initializer())
b3 = tf.Variable(tf.random_normal([512]))
L3 = tf.nn.relu(tf.matmul(L2, W3) + b3)
L3 = tf.nn.dropout(L3, keep_prob=keep_prob)
 
W4 = tf.get_variable("W4", shape=[512, 512],
                     initializer=tf.contrib.layers.xavier_initializer())
b4 = tf.Variable(tf.random_normal([512]))
L4 = tf.nn.relu(tf.matmul(L3, W4) + b4)
L4 = tf.nn.dropout(L4, keep_prob=keep_prob)
 
W5 = tf.get_variable("W5", shape=[512, 10],
                     initializer=tf.contrib.layers.xavier_initializer())
b5 = tf.Variable(tf.random_normal([10]))
hypothesis = tf.matmul(L4, W5) + b5
 
# define cost/loss & optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
    logits=hypothesis, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
 
# initialize
sess = tf.Session()
sess.run(tf.global_variables_initializer())
 
# train my model
for epoch in range(training_epochs):
    avg_cost = 0
    total_batch = int(mnist.train.num_examples / batch_size)
 
    for i in range(total_batch):
        batch_xs, batch_ys = mnist.train.next_batch(batch_size)
        feed_dict = {X: batch_xs, Y: batch_ys, keep_prob: 0.7}
        c, _ = sess.run([cost, optimizer], feed_dict=feed_dict)
        avg_cost += c / total_batch
 
    print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.9f}'.format(avg_cost))
 
print('Learning Finished!')
 
# Test model and check accuracy
correct_prediction = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print('Accuracy:', sess.run(accuracy, feed_dict={
      X: mnist.test.images, Y: mnist.test.labels, keep_prob: 1}))
 
# Get one and predict
r = random.randint(0, mnist.test.num_examples - 1)
print("Label: ", sess.run(tf.argmax(mnist.test.labels[r:r + 1], 1)))
print("Prediction: ", sess.run(
    tf.argmax(hypothesis, 1), feed_dict={X: mnist.test.images[r:r + 1], keep_prob: 1}))
 
plt.imshow(mnist.test.images[r:r + 1].
          reshape(28, 28), cmap='Greys', interpolation='nearest')
plt.show()
 
'''
Epoch: 0001 cost = 0.447322626
Epoch: 0002 cost = 0.157285590
Epoch: 0003 cost = 0.121884535
Epoch: 0004 cost = 0.098128681
Epoch: 0005 cost = 0.082901778
Epoch: 0006 cost = 0.075337573
Epoch: 0007 cost = 0.069752543
Epoch: 0008 cost = 0.060884363
Epoch: 0009 cost = 0.055276413
Epoch: 0010 cost = 0.054631256
Epoch: 0011 cost = 0.049675195
Epoch: 0012 cost = 0.049125314
Epoch: 0013 cost = 0.047231930
Epoch: 0014 cost = 0.041290121
Epoch: 0015 cost = 0.043621063
Learning Finished!
Accuracy: 0.9804
'''
 
cs


* More Deep & Dropout :


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
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
import random
import matplotlib.pylab as plt
 
tf.set_random_seed(777)
 
mnist = input_data.read_data_sets('../MNIST_data/', one_hot=True)
 
sess = tf.InteractiveSession()
 
X = tf.placeholder(tf.float32, [None, 784])
Y = tf.placeholder(tf.float32, [None, 10])
 
W1 = tf.get_variable("W1", shape=[784, 256]
                     , initializer=tf.contrib.layers.xavier_initializer())
W2 = tf.get_variable("W2", shape=[256, 256]
                     , initializer=tf.contrib.layers.xavier_initializer())
W3 = tf.get_variable("W3", shape=[256, 10]
                     , initializer=tf.contrib.layers.xavier_initializer())
 
b1 = tf.Variable(tf.zeros([256]))
b2 = tf.Variable(tf.zeros([256]))
b3 = tf.Variable(tf.zeros([10]))
 
dropout_rate = tf.placeholder(tf.float32)
_L1 = tf.nn.relu(tf.matmul(X, W1) + b1)
L1 = tf.nn.dropout(_L1, keep_prob=dropout_rate)
_L2 = tf.nn.relu(tf.matmul(L1, W2) + b2)
L2 = tf.nn.dropout(_L2, keep_prob=dropout_rate)
hypothesis = tf.matmul(L2, W3) + b3
 
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=hypothesis, labels=Y))
train = tf.train.AdamOptimizer(0.001).minimize(cost)
 
tf.global_variables_initializer().run()
 
for i in range(5500):  # 5500
    batch_xs, batch_ys = mnist.train.next_batch(100)
    train.run({X: batch_xs, Y: batch_ys, dropout_rate: 0.7})
    print("cost:", cost.eval({X: batch_xs, Y: batch_ys, dropout_rate: 0.7}))
 
correct_prediction = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(accuracy.eval({X: mnist.test.images, Y: mnist.test.labels, dropout_rate: 1}))
print(hypothesis.eval({X: mnist.test.images, Y: mnist.test.labels, dropout_rate: 1}))
 
r = random.randint(0, mnist.test.num_examples - 1)
print('Label:', sess.run(tf.argmax(mnist.test.labels[r:r + 1], 1)))
print('Prediction:', sess.run(tf.argmax(hypothesis, 1), {X: mnist.test.images[r:r + 1], dropout_rate: 1}))
print(mnist.test.images[r:r + 1])
 
plt.imshow(mnist.test.images[r:r + 1].reshape(28, 28)
           , cmap='Greys', interpolation='nearest')
plt.show()
 
cs






저작자표시

'Python 활용 딥러닝' 카테고리의 다른 글

CNN으로 꽃 이미지 예측하기  (0) 2018.12.20
CNN + Convolution Neural Network  (0) 2018.12.20
Deep Learning 학습방법(Layer 구성, Backpropagation, Activation function ReLU)  (0) 2018.12.19
Neural Nets & Deep learning, Neural Nets for XOR  (0) 2018.12.19
학습에 영향을 주는 요소들 (Learning Rate, Overfitting, Online Learning)  (0) 2018.12.19

+ Recent posts

티스토리툴바