1 #CNN 2 x = tf.placeholder(tf.float32,[None,input_node],name="x_input") 3 y_ = tf.placeholder(tf.float32,[None,output_node],name="y_output") 4 5 #input-->layer1 6 w_1 = tf.Variable(tf.truncted_normal([input_node,L1_node],stdev=0.5)) 7 b_1 = tf.Variable(tf.constant(0.1,shape=[L1_node])) 8 l_conv1 = tf.nn.relu(tf.matmul(x,w_1)+b_1,strides=[1,2,2,1]) 9 l_pool1 = tf.nn.max_pool(l_conv1,strides=[1,2,2,1],ksize = [1,2,2,1],padding='SAME') 10 11 #layer1-->layder2 12 w_2 = tf.Variable(tf.truncted_normal([L1_node,L2_node],stddev=0.5)) 13 b_2 = tf.Variable(tf.constant(0.1,shape=[L2_node])) 14 l_conv2 = tf.nn.relu(tf.matmul(l_pool1,w_2)+b_2) 15 l_pool2 = tf.nn.max_pool(l_conv2,strides=[1,2,2,1],ksize = [1,2,2,1],padding='SAME') 16 17 #layser2-->fc 18 w_3 = tf.Variable(tf.truncted_normal([L2_node,fc_node],stddev=0.5)) 19 b_3 = tf.Variable(tf.constant(0.1,shape=[fc_node])) 20 l_3 = tf.reshape(l_pool2,[-1,]) 21 fc_1 = tf.nn.relu(tf.matmul(l_3,w_3)+b_3) 22 23 #fc-->dropout 24 drop = tf.nn.dropout(fc_1,keep_prob) 25 26 #dropout-->softmax 27 w_4 = tf.Variable(tf.truncted_normal([fc_node,output_node],stddev=0.5)) 28 b_4 = tf.Variable(tf.constant(0.1,shape=[output_node])) 29 y = tf.nn.softmax(tf.matmul(drop,w_4)+b_4) 30 31 cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=tf.argmax(y,1),labels=tf.argmax(y_,1)) 32 cross_entropy_mean = tf.reduce_mean(cross_entropy) 33 loss = cross_entropy+reularation 34 35 train_step = tf.train.GradientDescentOptimizer(leraning_rate).minimize(loss)#以何种方式何种学习率去优化何种目标 36 37 correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1)) 38 accuracy = tf.reduce_mean(tf.cast(correct_predictiontf.float32)) 39 40 with tf.session() as sess: 41 tf.global_variable_initializer().run() 42 43 for i in range(max_steps): 44 sess.run(train_step,feed_dict={x:,y:}) 45 46 if i%1000 == 0: 47 validate_accu = sess.run(accuracy,feed_dict={x:x_val,y:y_val}) 48 49 test_accu = sess.run(accuracy,feed_dict = {x:x_test,y:y_dict}) 50 51 #RNN 52 input_size = 28(28*28 image) 53 hidden_size = 256 54 layer_num = 2 55 class_num =10 56 57 x = tf.placeholder(tf.float32,[None,784]) 58 y = tf.placeholder(tf.float32,[None,class_num]) 59 keep_prob = tf.placeholder(tf.float32) 60 61 x = tf.reshape(x,[-1,28,28]) 62 63 #一层lstm 64 lstm_layer = tf.contrib.rnn.BasicLSTMCell(num_units=hidden_size,forget_bias=1.0,..) 65 66 #添加dropout 67 lstm_layer = tf.contrib.rnn.DropoutWrrapper(lstm_layer,input_keep_prob =1.0,output_keep_prob=keep_prob) 68 69 #堆叠多层 70 mlstm = tf.contrib.rnn.MultiRNNCell([lstm_layer]*layer_sum,...) 71 72 init_state = mlstm.zero_state(batch_size,dtype=tf.float32) 73 74 output = mlstm(x) 75 76 #添加softmax层 77 w = tf.Variable(tf.truncted_normal([hidden_size,class_num],stddev=0.1),dtype=tf.float32) 78 b = tf.Variable(tf.constant(0.1,shape=[class_num]),dtype=tf.float32) 79 y_ = tf.nn.softmax(tf.matmul(output,w)+b) 80 81 cross_entropy = tf.reduce_mean(-y*tf.log(y_)) 82 train_step = tf.train.AdamOptimizer(learning_rate).minimize(cross_entropy) 83 84 correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1)) 85 #tf.argmax(input, axis=None, name=None, dimension=None)此函数是对矩阵按行或列计算最大值,0:按列，此处按行 86 accuracy = tf.reduce_mean(tf.cast(correct_prediction,'float'))#tf.cast():数据格式转换，此处bool-->float 87 88 with tf.Session as sess: 89 90 sess.run(train_step,feed_dict={x:,y:,keep_prob:}) #train 91 92 if i%1000 ==0: 93 train_accuracy = sess.run(accuracy,feed_dict={x:x_val,y:y_val,keep_prob:}) 94 print(train_accuracy) 95 96 #测试集 97 test_accuracy = sess.run(accuracy,feed_dict={x:x_test,y:y_test,keep_prob:}) 98 print(test_accuracy)