TensorFlow realization of convolution neural network (Simple) _ Neural network

Code (with detailed comments for source code) and dataset can be downloaded in github:
Https://github.com/crazyyanchao/TensorFlow-HelloWorld

#-*-Coding:utf-8-*-' convolution neural network test mnist data ' ######## #导入MNIST数据 ######## from Tensorflow.examples.tutorials.mnist Import input_data import TensorFlow as tf mnist = input_data.read_data_sets (' mnist_data/', one_hot=true) # Create default Interactiv Esession sess = tf.
InteractiveSession () ######## #卷积网络会有很多的权重和偏置需要创建, first define the initialization function in order to reuse ######## # make some random noises to the weights break the complete symmetry (such as the truncated normal distribution noise, the standard deviation set to 0.1) def weight_variable (shape): initial = Tf.truncated_normal (shape, stddev=0.1) return TF. Variable (initial) # because we want to use Relu, we also add some small positive values to the bias (0.1) To avoid death node (dead neurons) def bias_variable (shape): initial = Tf.constant (0.1, Shape=shape) return TF.  Variable (initial) ####### #卷积层, the pool layer is reused next, define the CREATE function ######## # tf.nn.conv2d is a 2-dimensional convolution function def TensorFlow in conv2d (x, W): return tf.nn.conv2d (x, W, strides=[1, 1, 1, 1], padding= ' SAME ') # Maximum pooled def 2*2 (x) with max_pool_2x2: Return Tf.nn.max_pool (x, KSI Ze=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding= ' SAME ') ####### #正式设计卷积神经网络之前先定义placeholder ######## # x is characteristic, Y_ is the real label. Convert picture data from 1D to 2D. Using tensor to transform functionsTf.reshape x = Tf.placeholder (Tf.float32, Shape=[none, 784]) Y_ = Tf.placeholder (Tf.float32, Shape=[none, ten]) X_image = t F.reshape (x,[-1,28,28,1]) ####### #设计卷积神经网络 ######## # First layer convolution # convolution kernel size for 5*5,1 color channel, 32 different convolution nuclei W_CONV1 = weight_variable ([5, 5 , 1, 32]) # using the CONV2D function for convolution operations, plus the bias b_conv1 = Bias_variable ([32]) # to convolution the x_image and weights vector, plus the offset, and then apply the Relu activation function, H_CONV1 = Tf.nn.rel U (conv2d (X_image, W_CONV1) + b_conv1) # Pool The output of the convolution h_pool1 = max_pool_2x2 (H_CONV1) # Second-tier convolution (roughly the same as the first layer, the convolution core is 64, this layer of convolution will extract 6 4 characteristics) = Weight_variable ([5, 5, w_conv2]) b_conv2 = Bias_variable ([)] H_conv2 = Tf.nn.relu (conv2d (H_pool1, W_conv2 + b_conv2) h_pool2 = max_pool_2x2 (h_conv2) # fully connected layer. Number of suppressed nodes 1024. Using the Relu activation function W_FC1 = weight_variable ([7 * 7 *, 1024]) B_FC1 = bias_variable ([1024]) 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) # In order to prevent the fitting, add the dropout layer before the output layer Keep_prob = Tf.placehold ER (tf.float32) h_fc1_drop = Tf.nn.dropout (H_FC1, keep_prob) # output layer. Add a Softmax layer, just like SOFTmax regression the same.
Get the probability output. W_FC2 = weight_variable ([1024, ten]) B_FC2 = Bias_variable ([ten]) Y_conv=tf.nn.softmax (Tf.matmul (H_fc1_drop, W_FC2) + b_ FC2) ####### #模型训练设置 ######## # defines loss function as cross entropy, the optimizer uses Adam, and gives a relatively small learning rate of 1e-4 cross_entropy = Tf.reduce_mean (-tf.reduce_sum (Y_*tf.log (Y_conv), reduction_indices=[1]) Train_step = Tf.train.AdamOptimizer (1e-4). Minimize ( Cross_entropy) # Defines the operation of the evaluation accuracy rate correct_prediction = tf.equal (Tf.argmax (y_conv,1), Tf.argmax (y_,1)) accuracy = Tf.reduce_ Mean (Tf.cast (correct_prediction, Tf.float32)) ####### #开始训练过程 ######## # initializes all Parameters Tf.global_variables_initializer (). Run () # Training (the kepp_prob ratio for dropout when setting up training is 0.5.) Mini-batch 50, 2000 iterations, participation in training samples 50,000 # Every 100 training, the accuracy of the evaluation Keep_prob set to 1, real-time monitoring model performance for I in range (1000): Batch = Mni St.train.next_batch (m) if i%100 = = 0:train_accuracy = Accuracy.eval (feed_dict={x:batch[0), Y_: batch[1], Keep_prob : 1.0}) print "-->step%d, training accuracy%.4f"% (i, train_accuracy) Train_step.run (feed_dict={x:bAtch[0], Y_: batch[1], keep_prob:0.5} # After full training, complete test on final Test set to get overall classification accuracy print "convolution neural network in mnist DataSet correct rate:%g"%
 Accuracy.eval (feed_dict={x:mnist.test.images, Y_: Mnist.test.labels, keep_prob:1.0})