#coding=utf-8

import input_data

import tensorflow as tf

import numpy as np

mnist = input_data.read_data_sets("MNIST_data/",one_hot=True)

learning_rate

init_func = tf.zeros

with tf.name_scope('Inputs'):

x_in = tf.placeholder(tf.float32,[None,784],name='x_in')

y_in = tf.placeholder(tf.float32,[None,10],name='y_in')

with tf.name_scope('input_reshape'):

image_shaped_input = tf.reshape(x_in, [-1, 28, 28, 1])

tf.summary.image('input_image',image_shaped_input, 10)

def def_var(rate=None,func = None):

global learning_rate

global init_func

global cross_entropy

global train_step

init_func = func

learning_rate = rate

train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(cross_entropy)

with tf.name_scope('Outputs'):

w_output = tf.Variable(init_func([784,10]),name='w_output')

tf.summary.histogram('w_output',w_output)

b_output = tf.Variable(init_func([1,10]),name='b_output')

tf.summary.histogram('b_output',b_output)

y_pre = tf.nn.softmax(tf.matmul(x_in,w_output)+b_output)

with tf.name_scope('Cross_entropy'):

cross_entropy = -tf.reduce_sum(y_in*tf.log(y_pre))

tf.summary.scalar('cross_entropy',cross_entropy)

with tf.name_scope('Train'):

train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(cross_entropy)

with tf.name_scope('Accuracy'):

correct_prediction = tf.equal(tf.argmax(y_pre,1), tf.argmax(y_in,1))

accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

tf.summary.scalar('accuracy',accuracy)

def make_hparam_string(rate=None,func=None):

if func == tf.zeros:

return '%s,tf.zeros'%(rate)

else:

return '%s,tf.random_normal'%(rate)

with tf.Session() as sess:

for rate in [1E-2,1E-3]:

for func in [tf.zeros,tf.random_normal]:

def_var(rate,func)

hparam_str = make_hparam_string(learning_rate,init_func)

writter = tf.summary.FileWriter('normal_MNIST/'+hparam_str,sess.graph)

merge_all = tf.summary.merge_all()

sess.run(tf.global_variables_initializer())

for i in range(1001):

'''

if i == 1:

print sess.run(w_output)

print sess.run(b_output)

print learning_rate

'''

batch_xs,batch_ys = mnist.train.next_batch(50)

if i % 50 == 0:

rs = sess.run(merge_all,feed_dict={x_in:batch_xs,y_in:batch_ys})

writter.add_summary(rs,i)

sess.run(train_step,feed_dict={x_in: batch_xs, y_in: batch_ys})

print '%s 精确度：%f' % (hparam_str,accuracy.eval(feed_dict={x_in: mnist.test.images, y_in: mnist.test.labels}))