我正在构建一个用于对MNIST数据进行分类的卷积神经网络。我使用了2个转换层和2个完全连接的层。
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
def _net_params():
weights = {
'conv1': tf.Variable(tf.random_normal([5, 5, 1, 32])),
'conv2': tf.Variable(tf.random_normal([5, 5, 32, 64])),
'fc1': tf.Variable(tf.random_normal([7 * 7 * 64, 1024])),
'fc2': tf.Variable(tf.random_normal([1024, 10])),
}
biases = {
'conv1': tf.Variable(tf.random_normal([32]),tf.float32),
'conv2': tf.Variable(tf.random_normal([64]),tf.float32),
'fc1': tf.Variable(tf.random_normal([1024]),tf.float32),
'fc2': tf.Variable(tf.random_normal([10]),tf.float32),
}
return weights, biases
def _fc_layer(inputs, weights, biases):
return tf.add(tf.matmul(inputs, weights), biases)
def _conv_layer(inputs, weights, biases, stride=1, padding='SAME'):
layer = tf.nn.conv2d(input=inputs,filter=weights,
strides=[1, stride, stride, 1],padding=padding)
layer = tf.nn.bias_add(layer, biases)
return tf.nn.relu(layer)
def pool_layer(inputs):
pool = tf.nn.max_pool(inputs, ksize=[1, 2, 2, 1], strides=[1, 2,
2, 1], padding="SAME")
return pool
def conv_net(x):
weights, biases = _net_params()
x = tf.reshape(x, shape=[-1, 28, 28, 1])
# Conv layers
conv1 = _conv_layer(x, weights['conv1'], biases['conv1'])
pool1 = pool_layer(conv1)
conv2 = _conv_layer(pool1, weights['conv2'], biases['conv2'])
pool2 = pool_layer(conv2)
flattened = tf.reshape(pool2, [-1, 7 * 7 * 64])
fc1 = _fc_layer(flattened, weights['fc1'], biases['fc1'])
fc1 = tf.nn.relu(fc1)
fc2 = _fc_layer(fc1, weights['fc2'], biases['fc2'])
return fc2
def _training():
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
learning_rate_ = tf.placeholder(tf.float32)
pred = conv_net(x)
cost = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=pred,
labels=y_))
optimizer = tf.train.AdamOptimizer(
learning_rate=learning_rate_).minimize(cost)
# optimizer = tf.train.GradientDescentOptimizer(
learning_rate=learning_rate_).minimize(cost)
correct = tf.equal(tf.argmax(pred, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
return x, y_, learning_rate_, optimizer, cost, accuracy
def main():
mnist = input_data.read_data_sets('tmp/data', one_hot=True)
n_epochs = 3
batch_size = 200
learning_rate = 0.005
x, y_, learning_rate_, optimizer, cost, accuracy = _training()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
current_epoch = 0
while current_epoch < n_epochs:
current_epoch += 1
print('epoch %s' % (current_epoch,))
current_batch = 1
while current_batch * batch_size <= len(
mnist.train.images):
current_batch += 1
batch_x, batch_y = mnist.train.next_batch(batch_size)
sess.run(fetches=optimizer, feed_dict={x: batch_x,
y_: batch_y, learning_rate_: learning_rate, })
if current_batch % 75 == 0:
loss, acc = sess.run([cost, accuracy], feed_dict=
{x: batch_x, y_: batch_y, learning_rate_: 0.,})
print(' batch %s: batch_loss=%s,
training_accuracy=%s' % (current_batch,
loss, acc,))
print('Training complete !')
print('Final accuracy is %s' % sess.run(accuracy, feed_dict=
{x: mnist.test.images, y_: mnist.test.labels,
learning_rate_: 0.}))
if __name__ == '__main__':
main()
（将此代码置于堆栈中时，可能会出现一些缩进错误）
当我使用AdamOptimizer时，我的准确率> 95％。 Accuracy for AdamOptimizer
但是当我使用GradientDescentOptimizer时，我的准确率达到了10％。 accuracy for GradientDescentOptimizer
如果我想使用GradientDescentOptimizer，你知道为什么我会得到这个较低的准确度以及如何解决这个问题。
由于
