利用Tensorflow构建CNN卷积神经网络,以mnist手写识别为例,步骤梳理如下:
1.定义权重函数
def weight_variable(shape):
# 使用正态分布填充变量,标准差设置为0.1
initial = tf.truncated_normal(shape, stddev=0.1)
#返回相应的变量
return tf.Variable(initial)
2.定义偏置函数
def bias_Variable(shape):
# 定义一个常量该常量全部由0.1组成,shape由参数shape指定
initial = tf.constant(0.1, shape=shape)
# 返回一个变量,变量的值由刚刚定义的常量决定
return tf.Variable(initial)
3.定义卷积层
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
4.定义池化层
def max_pool(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
5.定义CNN网络结构
# 构建网络,由2个卷积层(包含激活层、池化层),一个全连接层,一个dropout层,一个softmax层 组成
def deepnn(x):
x_image = tf.reshape(x, [-1, 28, 28, 1])
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1)+b_conv1)
h_pool1 = max_pool(h_conv1)
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2)+b_conv2)
h_pool2 = max_pool(h_conv2)
W_fc1 = weight_variable([7*7*64, 1024])
b_fc1 = bias_variable([1024])
# reshape成向量
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1)+b_fc1)
keep_prob = tf.placeholder(tf.float32)
# dropout层
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
# softmax层
y_predict = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2)+b_fc2)
return keep_prob, y_predict
6.定义损失函数、参数优化器
# 定义交叉熵函数,即我们使用的loss函数,我们的目的就是在训练过程中使得loss函数尽可能的小。
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_actual*tf.log(y_predict), reduction_indices=[1]))
# 使用AdamOptimizer进行loss函数的优化。使用ADAM优化算法,它对每个权值都计算自适应的学习速率,收敛速度比一般的梯度下降法更快。
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
7.定义预测函数、准确率验证
# 定义预测函数,这里比较预测结果和实际结果的相匹配的数量
correct_prediction = tf.equal(tf.argmax(y_predict, 1), tf.argmax(y_actual, 1))
# 精确度计算
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
8.连接session,训练模型
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
for i in range(20000):
# 每次50张图片
batch = mnist.train.next_batch(50)
# 每训练100次,验证一次
if i%100 ==0:
train_acc = accuracy.eval(feed_dict={x:batch[0], y_actual: batch[1], keep_prob:1.0})
print 'step %d, training accuracy %g' % (i, train_acc)
train_step.run(feed_dict={x:batch[0], y_actual:batch[1], keep_prob:0.5})
test_acc = accuracy.eval(feed_dict={x: mnist.test.images, y_actual:mnist.test.labels, keep_prob:1.0})
print "test accuracy %g" % test_acc
由于Tensorflow的后台计算是依赖于高效的C++,因此我们需要连接到后台去运行。与后台的连接称为一个session会话。也就是说,前面搭建CNN的过程其实就是先构建一个graph图,然后在session里运行它。这就像我们家里要用自来水,就必须先接好水管,然后自来水才能在水管里运行流动。