深度学习之卷积神经网络CNN

import pickle import numpy as np import tensorflow as tf import random def load_data(data_path): with open(data_path,"rb") as f: u = pickle.Unpickler(f,encoding="latin1")#使用Pickler实例在序列化对象的时候， #从文件对象中读取一个序列化后的对象并将其解序列化成原来的层级结构 target_params = u.load() data = target_params['data'].reshape([-1,32,32,3]) label = target_params['labels'] label=np.array(label) return data,label data1,label1=load_data("D:/python3/lpthw/train/data_batch_1") data2,label2=load_data("D:/python3/lpthw/train/data_batch_2") data3,label3=load_data("D:/python3/lpthw/train/data_batch_3") data4,label4=load_data("D:/python3/lpthw/train/data_batch_4") data5,label5=load_data("D:/python3/lpthw/train/data_batch_5") data=np.concatenate((data1,data2,data3,data4,data5),axis=0) data=np.array(data, dtype=np.float32) #print(data.dtype) labels=np.concatenate((label1,label2,label3,label4,label5),axis=0) data=data/255.0 def load_data1(data_path): with open(data_path,"rd") as f: u=pickle.Unpickler(f,encoding="latin1") target_params=u.load() data=target_params['data'].reshape([-1,32,32,3]) label=target_params['labels'] label=np.array(label) return data,label test_data,test_label=load_data("D:/python3/lpthw/test/test_batch") test_data=np.array(test_data,dtype=np.float32) test_data=test_data/255.0 #Training parameters learning_rate=0.001 num_steps=10000 batch_size=100 num_input=3072 num_classes=10 dropout=0.1 traing_epoch=50000/batch_size X=tf.placeholder("float",[None,32,32,3]) Y=tf.placeholder("float",[None]) display_step=1 #卷积部分 def conv_net(X,n_classes,dropout,reuse,is_training): with tf.variable_scope('ConvNet',reuse=tf.AUTO_REUSE):#通过tf.variable_scope函数可以控制tf.get_variable函数的语义 conv1=tf.layers.conv2d(X,64,5,activation=tf.nn.relu)#filters,kernel conv1=tf.layers.max_pooling2d(conv1,2,2)#strides,kernel conv2=tf.layers.conv2d(conv1,64,5,activation=tf.nn.relu) conv2=tf.layers.max_pooling2d(conv2,2,2) fc1=tf.contrib.layers.flatten(conv2) fc1=tf.layers.dense(fc1,1024)#dense为全连接层，即add_layer函数，该层的输入，输出的维度 fc1=tf.layers.dropout(fc1,rate=dropout,training=is_training) out=tf.layers.dense(fc1,n_classes) return out logits_train=conv_net(X,num_classes,dropout,reuse=False,is_training=True)#train输出部分 prediction=tf.argmax(logits_train,axis=1) loss_op=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits_train,labels=tf.cast(Y,dtype=tf.int32))) train_op=tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss_op) accuracy= tf.metrics.accuracy(labels=Y, predictions=prediction) init=tf.global_variables_initializer() with tf.Session() as sess: sess.run(init) sess.run(tf.local_variables_initializer()) for step in range(0,num_steps): for j in range(0,int(traing_epoch)): batch_x=data[j*batch_size:j*batch_size+100] batch_y=labels[j*batch_size:j*batch_size+100] sess.run(train_op,feed_dict={X:batch_x,Y:batch_y}) if step % display_step ==0 or step ==1: loss,acc=sess.run([loss_op,accuracy],feed_dict={X:batch_x,Y:batch_y}) print("Epoch:",step,loss,acc) print("Testing Accuracy:",sess.run(accuracy,feed_dict={X:test_data,Y:test_label}))