Skip to content

Commit 190bb8d

Browse files
aymericdamienaymericdamien
committed
added alexnet & update loss
1 parent e43ca92 commit 190bb8d

2 files changed

Lines changed: 102 additions & 5 deletions

File tree

‎alexnet.py‎

Lines changed: 99 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,99 @@
1+
# Import MINST data
2+
import input_data
3+
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)
4+
5+
import tensorflow as tf
6+
7+
# Parameters
8+
learning_rate = 0.001
9+
training_iters = 200000
10+
batch_size = 64
11+
display_step = 20
12+
13+
#Network Parameters
14+
n_input = 784 #MNIST data input
15+
n_classes = 10 #MNIST total classes
16+
dropout = 0.8
17+
18+
# Create model
19+
x = tf.placeholder(tf.types.float32, [None, n_input])
20+
y = tf.placeholder(tf.types.float32, [None, n_classes])
21+
keep_prob = tf.placeholder(tf.types.float32) #dropout
22+
23+
def conv2d(name, l_input, w, b):
24+
return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(l_input, w, strides=[1, 1, 1, 1], padding='SAME'),b), name=name)
25+
26+
def max_pool(name, l_input, k):
27+
return tf.nn.max_pool(l_input, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='SAME', name=name)
28+
29+
def norm(name, l_input, lsize=4):
30+
return tf.nn.lrn(l_input, lsize, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name=name)
31+
32+
def conv_net(_X, _weights, _biases, _dropout):
33+
_X = tf.reshape(_X, shape=[-1, 28, 28, 1])
34+
35+
conv1 = conv2d('conv1', _X, _weights['wc1'], _biases['bc1'])
36+
pool1 = max_pool('pool1', conv1, k=2)
37+
norm1 = norm('norm1', pool1, lsize=4)
38+
norm1 = tf.nn.dropout(norm1, _dropout)
39+
40+
conv2 = conv2d('conv2', norm1, _weights['wc2'], _biases['bc2'])
41+
pool2 = max_pool('pool2', conv2, k=2)
42+
norm2 = norm('norm2', pool2, lsize=4)
43+
norm2 = tf.nn.dropout(norm2, _dropout)
44+
45+
conv3 = conv2d('conv3', norm2, _weights['wc3'], _biases['bc3'])
46+
pool3 = max_pool('pool3', conv3, k=2)
47+
norm3 = norm('norm3', pool3, lsize=4)
48+
norm3 = tf.nn.dropout(norm3, _dropout)
49+
50+
dense1 = tf.reshape(norm3, [-1, _weights['wd1'].get_shape().as_list()[0]])
51+
dense1 = tf.nn.relu(tf.matmul(dense1, _weights['wd1']) + _biases['bd1'], name='fc1')
52+
53+
dense2 = tf.nn.relu(tf.matmul(dense1, _weights['wd2']) + _biases['bd2'], name='fc2')
54+
55+
out = tf.matmul(dense2, _weights['out']) + _biases['out']
56+
return out
57+
58+
weights = {
59+
'wc1': tf.Variable(tf.random_normal([3, 3, 1, 64])),
60+
'wc2': tf.Variable(tf.random_normal([3, 3, 64, 128])),
61+
'wc3': tf.Variable(tf.random_normal([3, 3, 128, 256])),
62+
'wd1': tf.Variable(tf.random_normal([4*4*256, 1024])),
63+
'wd2': tf.Variable(tf.random_normal([1024, 1024])),
64+
'out': tf.Variable(tf.random_normal([1024, 10]))
65+
}
66+
67+
biases = {
68+
'bc1': tf.Variable(tf.random_normal([64])),
69+
'bc2': tf.Variable(tf.random_normal([128])),
70+
'bc3': tf.Variable(tf.random_normal([256])),
71+
'bd1': tf.Variable(tf.random_normal([1024])),
72+
'bd2': tf.Variable(tf.random_normal([1024])),
73+
'out': tf.Variable(tf.random_normal([n_classes]))
74+
}
75+
76+
pred = conv_net(x, weights, biases, keep_prob)
77+
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred, y))
78+
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
79+
80+
#Evaluate model
81+
correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
82+
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.types.float32))
83+
84+
# Train
85+
init = tf.initialize_all_variables()
86+
with tf.Session() as sess:
87+
sess.run(init)
88+
step = 1
89+
while step * batch_size < training_iters:
90+
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
91+
sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys, keep_prob: dropout})
92+
if step % display_step == 0:
93+
acc = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.})
94+
loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.})
95+
print "Iter " + str(step*batch_size) + ", Loss= " + "{:.6f}".format(loss) + ", Training Accuracy= " + "{:.5f}".format(acc)
96+
step += 1
97+
print "Optimization Finished!"
98+
#Accuracy on 256 mnist test images
99+
print "Testing Accuracy:", sess.run(accuracy, feed_dict={x: mnist.test.images[:256], y: mnist.test.labels[:256], keep_prob: 1.})

‎convolutional_network.py‎

Lines changed: 3 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -21,7 +21,7 @@
2121
keep_prob = tf.placeholder(tf.types.float32) #dropout
2222

2323
def conv2d(img, w, b):
24-
return tf.nn.relu(tf.nn.conv2d(img, w, strides=[1, 1, 1, 1], padding='SAME') + b)
24+
return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(img, w, strides=[1, 1, 1, 1], padding='SAME'),b))
2525

2626
def max_pool(img, k):
2727
return tf.nn.max_pool(img, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='SAME')
@@ -67,18 +67,16 @@ def conv_net(_X, _weights, _biases, _dropout):
6767
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.types.float32))
6868

6969
# Train
70-
#load mnist data
7170
init = tf.initialize_all_variables()
7271
with tf.Session() as sess:
7372
sess.run(init)
7473
step = 1
75-
avg_cost = 0.
7674
while step * batch_size < training_iters:
7775
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
7876
sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys, keep_prob: dropout})
7977
if step % display_step == 0:
80-
avg_cost += sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.})/batch_size
81-
print "Iter", str(step*batch_size), "cost=", "{:.9f}".format(avg_cost/step)
78+
loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keep_prob: dropout})/batch_size
79+
print "Iter", str(step*batch_size), "loss=", "{:.9f}".format(loss/step)
8280
step += 1
8381
print "Optimization Finished!"
8482
#Accuracy on 256 mnist test images

0 commit comments

Comments
 (0)