#coding=utf-8
import tensorflow  as tf
import numpy as np
import matplotlib .pyplot  as plt
from tensorflow .examples .tutorials .mnist  import  input_data



#define dataset

mnist=input_data .read_data_sets ("/home/nvidia/Downloads/",one_hot= True )


#defien agruments


batch_zize=20
iter=np.int(mnist .train.images.shape[0]/batch_zize )
print(iter )


#define learning_rate

LEARNING_RATE_STEP=100
LEARNING_RATE_BASE=0.001
LEARNING_RATE_DECAY=0.99
global_step=tf.Variable (0,trainable= False )
learning_rate=tf.train.exponential_decay (learning_rate= LEARNING_RATE_BASE ,global_step= global_step ,decay_steps= LEARNING_RATE_STEP
                                          ,decay_rate= LEARNING_RATE_DECAY ,staircase= True )



#define tool

def Weight_V(shape):
     weight=tf.truncated_normal (shape=shape,stddev= 0.1)
     return tf.Variable (weight )


def bias_V(shape):
     bia_=tf.constant (shape=shape,value= 0.1)
     return tf.Variable (bia_ )


def conv2d_(x,w):
     return tf.nn.conv2d (x,filter= w,padding= "SAME",strides= [1,1,1,1])


def max_pool(x):
     return tf.nn.max_pool (x,ksize= [1,2,2,1],strides=[1,2,2,1],padding="SAME")



#define net


x_input=tf.placeholder (shape=[None,784],dtype= tf.float32)
y_input=tf.placeholder (shape= [None,10],dtype= tf.float32)



x =tf.reshape(x_input ,shape= [-1,28,28,1])



#
w_conv1=Weight_V(shape= [5,5,1,32])
b_conv1=bias_V(shape= [32])
c_conv1=tf.nn.relu (conv2d_(x ,w_conv1 )+b_conv1  )
m_conv1=max_pool(c_conv1 )
#14*14*32


w_conv2=Weight_V(shape= [5,5,32,64])
b_conv2=bias_V(shape= [64])
c_conv2=tf.nn.relu (conv2d_(m_conv1 ,w_conv2 )+b_conv2  )
m_conv2=max_pool(c_conv2 )
#7*7*64


w_fc1=Weight_V([7*7*64,1024])
b_fc1=bias_V(shape= [1024])
c_fc1=tf.reshape(m_conv2 ,[-1,7*7*64])
fc1=tf.nn.relu(tf.matmul(c_fc1 ,w_fc1 )+b_fc1 )



w_fc2=Weight_V(shape= [1024,10])
b_fc2=bias_V(shape= [10])
prediction=tf.nn.softmax (tf.matmul(fc1,w_fc2 )+b_fc2 )


#define

# correct_accurcy=tf.equal(tf.argmax(prediction,axis=1),tf.argmax(y_input,axis=1))
# accurcy=tf.reduce_mean(tf.cast(correct_accurcy,dtype=tf.float32))

correct_accurcy=tf.equal (tf.argmax (prediction ,axis= 1),tf.argmax (y_input ,axis= 1))

accurcy=tf.reduce_mean (tf.cast(correct_accurcy ,dtype= tf.float32))



#traing backward
#
crosss_entropy =-tf.reduce_mean (y_input *tf.log(prediction ))
train_step=tf.train.GradientDescentOptimizer (learning_rate).minimize(crosss_entropy,global_step= global_step  )




#initial global argumnets

init=tf.global_variables_initializer ()


#SESS

with tf.Session() as sess:
     sess.run(init)
     for i in range(21):
          X,Y=mnist .test.next_batch(100)
          for j in range(iter ):
               xt,yt=mnist .train.next_batch (batch_zize )
               sess.run(train_step ,feed_dict= {x_input :xt,y_input :yt})


          acc=sess.run(accurcy ,feed_dict= {x_input :X,y_input :Y})
          print(acc)