tensorflow入门02——搭建逻辑回归模型

搭建逻辑回归模型对mnist数据集中手写字体进行分类

1、加载数据

minst=input_data.read_data_sets('data/',one_hot=True)
trainimg=minst.train.images
trainlabel=minst.train.labels
testimg=minst.test.images
testlabel=minst.test.labels
print('minst loaded')

2、设置变量

x=tf.placeholder("float",[None,784])
y=tf.placeholder("float",[None,10])
W=tf.Variable(tf.zeros([784,10]))
b=tf.Variable(tf.zeros([10]))

3、回归模型

这里回归模型用softmax做十分类任务

#logistic regression model
actv=tf.nn.softmax(tf.matmul(x,W)+b)
#cost
cost=tf.reduce_mean(-tf.reduce_sum(y*tf.log(actv),reduction_indices=1))
#optimizer
learning_rate=0.01
optm=tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cost)
#prediction
pred=tf.equal(tf.argmax(actv,1), tf.argmax(y,1))
#accuracy
accr=tf.reduce_mean(tf.cast(pred,"float"))

4、开始训练

#initializer
init=tf.global_variables_initializer()
sess=tf.InteractiveSession()

training_epochs=50
batch_size=100
display_step=5

sess=tf.Session()
sess.run(init)

for epoch in range(training_epochs):
    avg_cost=0.
    num_batch=int(minst.train.num_examples/batch_size)
    for i in range(num_batch):
        batch_xs,batch_ys=minst.train.next_batch(batch_size)
        feeds={x:batch_xs,y:batch_ys}
        sess.run(optm,feed_dict=feeds)
        avg_cost+=sess.run(cost,feeds)/num_batch
        #display
        if epoch%display_step==0:
            feeds_train={x:batch_xs,y:batch_ys}
            feeds_test={x:minst.test.images,y:minst.test.labels}
            train_acc=sess.run(accr,feed_dict=feeds_train)
            test_acc=sess.run(accr,feed_dict=feeds_test)
            print("Epoch:%03d/%03d cost:%.9f train_accr:%.3f test_accr:%.3f"%(epoch,training_epochs,avg_cost,train_acc,test_acc))
print("Done")

5、训练结果