Tensorflow实现逻辑回归模型

import tensorflow as tf
##import input_data
from tensorflow.examples.tutorials.mnist import  input_data

##下载导入mnist数据集 如果出现连接问题 请连vpn
mnist = input_data.read_data_sets('data/', one_hot=True)

##part1  初始化参数
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]))

##part2 定义函数
actv = tf.nn.softmax(tf.matmul(x, W) + b)
cost = tf.reduce_mean(-tf.reduce_sum(y * tf.log(actv), reduction_indices=1))
learning_rate = 0.01
optm = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)

pred = tf.equal(tf.arg_max(actv, 1), tf.arg_max(y, 1))
accr=tf.reduce_mean(tf.cast(pred,'float'))

##part3 迭代运算
training_epochs=50
batch_size=100
display_step=5

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

for  epoch in range(1,(training_epochs+1)): ##[1~50]
    avg_cost=0
    num_batch=int(mnist.train.num_examples/batch_size)
    for i  in range(num_batch):
        batch_xs,batch_ys=mnist.train.next_batch(batch_size)
        feeds = {x: batch_xs, y: batch_ys}
        sess.run(optm,feed_dict=feeds)
        avg_cost+=sess.run(cost,feed_dict=feeds)/num_batch
    if epoch%display_step==0:
        ##feeds_train={x:batch_xs,y:batch_ys}
        feeds_test = {x: mnist.test.images, y: mnist.test.labels}
        train_acc=sess.run(accr,feed_dict=feeds)
        test_acc=sess.run(accr,feed_dict=feeds_test)
        print("Epoch:%03d/%03d cost:%.9f train_acc:%.3f test_acc:%.3f"%(epoch,training_epochs,avg_cost,train_acc,test_acc))