I'm following this tutorial to learn tensorflow and tensorboard. Below are my code. The accuracy is stuck around random. I couldn't find out where is wrong.
Can someone point out where the bug is? I would also like to know how one should debug in tensorflow. Thanks.
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
import tensorflow as tf
def conv_layer(input, size_in, size_out, name="conv"):
with tf.name_scope(name):
w = tf.Variable(tf.truncated_normal([5, 5, size_in, size_out], stddev=0.1), name="W")
b = tf.Variable(tf.constant(0.1, shape=[size_out]), name="B")
conv = tf.nn.conv2d(input, w, strides=[1,1,1,1], padding="SAME")
act = tf.nn.relu(conv + b)
tf.summary.histogram("weights", w)
tf.summary.histogram("biases", b)
tf.summary.histogram("activations", act)
return tf.nn.max_pool(act, ksize=[1,2,2,1], strides=[1,2,2,1], padding="SAME")
def fc_layer(input, size_in, size_out, name="fc"):
with tf.name_scope(name):
w = tf.Variable(tf.truncated_normal([size_in, size_out], stddev=0.1), name="W")
b = tf.Variable(tf.constant(0.1, shape=[size_out]), name="B")
act = tf.nn.relu(tf.matmul(input, w) + b)
tf.summary.histogram("weights", w)
tf.summary.histogram("biases", b)
tf.summary.histogram("activations", act)
return act
def mnist_model(learning_rate, path):
tf.reset_default_graph()
sess = tf.Session()
x = tf.placeholder(tf.float32, shape=[None, 784], name="x")
x_image = tf.reshape(x, [-1, 28, 28, 1])
tf.summary.image('input', x_image, 3)
y = tf.placeholder(tf.float32, shape=[None, 10], name="labels")
conv1 = conv_layer(x_image, 1, 32, "conv1")
conv_out = conv_layer(conv1, 32, 64, "conv2")
flattened = tf.reshape(conv_out, [-1, 7 * 7 * 64])
fc1 = fc_layer(flattened, 7 * 7 * 64, 1024, "fc1")
logits = fc_layer(fc1, 1024, 10, "fc2")
with tf.name_scope("xent"):
xent = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(
logits=logits, labels=y), name="xent")
tf.summary.scalar("xent", xent)
with tf.name_scope("train"):
train_step = tf.train.AdamOptimizer(learning_rate).minimize(xent)
with tf.name_scope("accuracy"):
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar("accuracy", accuracy)
summ = tf.summary.merge_all()
sess.run(tf.global_variables_initializer())
writer = tf.summary.FileWriter(path)
writer.add_graph(sess.graph)
for i in range(2000):
batch = mnist.train.next_batch(100)
if i % 50 == 0:
[train_accuracy, s] = sess.run([accuracy, summ], feed_dict={x: batch[0], y: batch[1]})
print train_accuracy
writer.add_summary(s, i)
sess.run(train_step, feed_dict={x: batch[0], y: batch[1]})
mnist_model(1e-3, path = "/tmp/mnist_demo/10")
0.09
0.08
0.04
0.07
0.12
0.12
0.09
0.12
0.08
0.1
0.11
0.14
0.11
0.11
0.13
0.11
0.19
0.06
The problem is you apply relu activation on the last layer, so all the logits are thresholded at zero.
Solution:
change
def fc_layer(input, size_in, size_out, name="fc"):
with tf.name_scope(name):
w = tf.Variable(tf.truncated_normal([size_in, size_out], stddev=0.1), name="W")
b = tf.Variable(tf.constant(0.1, shape=[size_out]), name="B")
act = tf.nn.relu(tf.matmul(input, w) + b)
tf.summary.histogram("weights", w)
tf.summary.histogram("biases", b)
tf.summary.histogram("activations", act)
return act
to
def fc_layer(input, size_in, size_out, name="fc", activation=tf.nn.relu):
with tf.name_scope(name):
w = tf.Variable(tf.truncated_normal([size_in, size_out], stddev=0.1), name="W")
b = tf.Variable(tf.constant(0.1, shape=[size_out]), name="B")
act = tf.matmul(input, w) + b
if activation is not None:
act = activation(act)
tf.summary.histogram("weights", w)
tf.summary.histogram("biases", b)
tf.summary.histogram("activations", act)
return act
And pass None as activation in the last fully-connected layer:
logits = fc_layer(fc1, 1024, 10, "fc2", activation=None)