-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel.py
128 lines (103 loc) · 5.5 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
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import functions
from tensorflow.python.framework import ops
train_X_orig,train_y_orig,test_X_orig,test_y_orig = load_mnist_data()
train_X=train_X_orig.reshape(-1,28,28,1)/255
test_X=test_X_orig.reshape(-1,28,28,1)/255
train_y=convert_to_one_hot(train_y_orig,10)
test_y=convert_to_one_hot(test_y_orig,10)
def model(train_X,train_y,test_X,test_y,learning_rate,epoch,batch_size):
ops.reset_default_graph()
(m,nx1,nx2,_)=train_X.shape
n_y=train_y.shape[0]
X,Y=create_placeholders(nx1,nx2,n_y)
costs=[]
channels=np.array([1,16,32,64])
krnls=initialize_kernels(channels)
logits=network(X,krnls)
loss=tf.losses.softmax_cross_entropy(onehot_labels=Y,logits=logits)
optimizer=tf.train.AdamOptimizer(learning_rate).minimize(loss)
init=tf.global_variables_initializer()
saver=tf.train.Saver()
with tf.Session() as sess:
sess.run(init)
for epoch in range(epoch):
epoch_cost=0
minibatch_cost=0
num_mini=int(m/batch_size)
minibatch=random_mini_batches(train_X,train_y,100)
for mini in minibatch:
(mini_X,mini_y)=mini
_,minibatch_cost=sess.run([optimizer,loss],feed_dict={X:mini_X,Y:mini_y})
epoch_cost=epoch_cost+minibatch_cost/num_mini
# if epoch%100==0:
print("Cost after epoch %i: %f" %(epoch,epoch_cost))
#saver.save(sess,"E:/spider/Fashion_mnist/mnist.ckpt")
if epoch%5==0:
costs.append(epoch_cost)
plt.plot(np.squeeze(costs))
plt.ylabel('cost')
plt.xlabel('iterations (per tens)')
plt.title("Learning rate =" + str(learning_rate))
plt.show()
correct_prediction = tf.equal(tf.argmax(logits,axis=1), tf.argmax(Y,axis=1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print ("Train Accuracy:", accuracy.eval({X: train_X, Y: train_y}))
print ("Test Accuracy:", accuracy.eval({X: test_X, Y: test_y}))
saver.save(sess,"E:/spider/Fashion_mnist/mnist.ckpt")
return None
def network(X,krnls):
inputs=tf.cast(tf.reshape(X,[-1,28,28,1]),tf.float32)
conv1=tf.nn.conv2d(inputs,krnls["1"],padding="SAME",strides=[1,1,1,1],name='conv1')
pool1=tf.layers.max_pooling2d(inputs=conv1,pool_size=[2,2],strides=[2,2],name='pool1')
b_norm1=tf.nn.relu(tf.layers.batch_normalization(inputs=pool1,momentum=0.99,center=True,scale=True,epsilon=0.000001,name="b_norm1"))
# conv1=tf.layers.conv2d(inputs,filters=32,kernel_size=[3,3],strides=1,padding='same',activation='relu')
# conv2=tf.layers.conv2d(inputs=pool1,filters=64,kernel_size=[3,3],strides=1,padding='same',activation='relu')
conv2=tf.nn.conv2d(b_norm1,krnls["2"],padding="SAME",strides=[1,1,1,1],name='conv2')
pool2=tf.layers.max_pooling2d(inputs=conv2,pool_size=[2,2],strides=[2,2],name='pool2')
b_norm2=tf.nn.relu(tf.layers.batch_normalization(inputs=pool2,momentum=0.99,center=True,scale=True,epsilon=0.000001,name="b_norm2"))
# conv3=tf.layers.conv2d(inputs=pool2,filters=128,kernel_size=[3,3],strides=1,padding='same',activation='relu')
conv3=tf.nn.conv2d(b_norm2,krnls["3"],padding="SAME",strides=[1,1,1,1],name='conv3')
pool3=tf.layers.max_pooling2d(inputs=conv3,pool_size=[2,2],strides=[2,2],name='pool3')
b_norm3=tf.layers.batch_normalization(inputs=pool3,momentum=0.99,center=True,scale=True,epsilon=0.000001,name="b_norm3")
flat=tf.reshape(pool3,[-1,pool3.shape[1]*pool3.shape[2]*pool3.shape[3]])
dense1=tf.layers.dense(inputs=flat,units=512,activation=None,name="dense1")
b_norm4=tf.nn.relu(tf.layers.batch_normalization(inputs=dense1,momentum=0.99,scale=True,center=True,epsilon=0.000001,name="b_norm4"))
dropout1=tf.layers.dropout(inputs=b_norm4,rate=0.5,name="dropout1")
dense2=tf.layers.dense(inputs=dropout1,units=256,activation=None,name="dense2")
b_norm5=tf.nn.relu(tf.layers.batch_normalization(inputs=dense2,momentum=0.99,scale=True,center=True,epsilon=0.000001,name="b_norm5"))
dropout2=tf.layers.dropout(inputs=b_norm5,rate=0.5,name="dropout2")
logits=tf.layers.dense(inputs=dropout2,units=10,name="dense3")
return logits
def create_placeholders(w,h,ny):
x=tf.placeholder(tf.float32,[None,w,h,None],name='x')
y=tf.placeholder(tf.int32,[None,10],name='y')
return x,y
def initialize_kernels(channels):
L=len(channels)
parameter=dict()
for i in range(1,L):
parameter[str(i)]=tf.get_variable(name=str(i),shape=[3,3,channels[i-1],channels[i]],dtype=tf.float32,initializer=tf.contrib.layers.xavier_initializer())
#parameter["b"+str(i)]=tf.get_variable(name="b"+str(i),shape=[channels[i]],dtype=tf.float32,initializer=tf.zeros_initializer())
print(parameter[str(i)].shape)
return parameter
def predict(X1,Y1):
tf.reset_default_graph()
(nx1,nx2,c)=X1.shape
n_y=Y1.shape[0]
X,Y=create_placeholders(nx1,nx2,n_y)
costs=[]
channels=np.array([1,16,32,64])
krnls=initialize_kernels(channels)
logits=network(X,krnls)
saver=tf.train.Saver()
sess=tf.Session()
saver.restore(sess,"E:/spider/Fashion_mnist/mnist.ckpt")
p=tf.argmax(logits,axis=1)
prediction=sess.run(p,feed_dict={X:X1.reshape(-1,28,28,1),Y:Y1.reshape(-1,10)})
plt.imshow(X1.reshape(28,28))
print("Predicted class:",prediction," Original class:",np.argmax(Y1,axis=1))
return None
model(train_X,train_y,test_X,test_y,0.001,40,600)