#!/usr/bin/env python2# -*- coding: utf-8 -*-"""@author"""

#import modulesimport numpy as npimport matplotlib.pyplot as plt#from sklearn.metrics import confusion_matriximport tensorflow as tfimport timefrom datetime import timedeltaimport mathfrom tensorflow.examples.tutorials.mnist import input_data

def new_weights(shape):  return tf.Variable(tf.truncated_normal(shape,stddev=0.05))def new_biases(length):  return tf.Variable(tf.constant(0.1,shape=length))def conv2d(x,W):  return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME')def max_pool_2x2(inputx):  return tf.nn.max_pool(inputx,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')

#import datadata = input_data.read_data_sets("./data", one_hot=True) # one_hot means [0 0 1 0 0 0 0 0 0 0] stands for 2

print("Size of:")print("--Training-set:\t\t{}".format(len(data.train.labels)))print("--Testing-set:\t\t{}".format(len(data.test.labels)))print("--Validation-set:\t\t{}".format(len(data.validation.labels)))data.test.cls = np.argmax(data.test.labels,axis=1)  # show the real test labels: [7 2 1 ..., 4 5 6], 10000values

x = tf.placeholder("float",shape=[None,784],name='x')x_image = tf.reshape(x,[-1,28,28,1])

y_true = tf.placeholder("float",shape=[None,10],name='y_true')y_true_cls = tf.argmax(y_true,dimension=1)# Conv 1layer_conv1 = {"weights":new_weights([5,5,1,32]),        "biases":new_biases([32])}h_conv1 = tf.nn.relu(conv2d(x_image,layer_conv1["weights"])+layer_conv1["biases"])h_pool1 = max_pool_2x2(h_conv1)# Conv 2layer_conv2 = {"weights":new_weights([5,5,32,64]),        "biases":new_biases([64])}h_conv2 = tf.nn.relu(conv2d(h_pool1,layer_conv2["weights"])+layer_conv2["biases"])h_pool2 = max_pool_2x2(h_conv2)# Full-connected layer 1fc1_layer = {"weights":new_weights([7*7*64,1024]),      "biases":new_biases([1024])}h_pool2_flat = tf.reshape(h_pool2,[-1,7*7*64])h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat,fc1_layer["weights"])+fc1_layer["biases"])# Droupout Layerkeep_prob = tf.placeholder("float")h_fc1_drop = tf.nn.dropout(h_fc1,keep_prob)# Full-connected layer 2fc2_layer = {"weights":new_weights([1024,10]),       "biases":new_weights([10])}# Predicted classy_pred = tf.nn.softmax(tf.matmul(h_fc1_drop,fc2_layer["weights"])+fc2_layer["biases"]) # The output is like [0 0 1 0 0 0 0 0 0 0]y_pred_cls = tf.argmax(y_pred,dimension=1) # Show the real predict number like '2'# cost function to be optimizedcross_entropy = -tf.reduce_mean(y_true*tf.log(y_pred))optimizer = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(cross_entropy)# Performance Measurescorrect_prediction = tf.equal(y_pred_cls,y_true_cls)accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))with tf.Session() as sess:  init = tf.global_variables_initializer()  sess.run(init)  train_batch_size = 50  def optimize(num_iterations):    total_iterations=0    start_time = time.time()    for i in range(total_iterations,total_iterations+num_iterations):      x_batch,y_true_batch = data.train.next_batch(train_batch_size)      feed_dict_train_op = {x:x_batch,y_true:y_true_batch,keep_prob:0.5}      feed_dict_train = {x:x_batch,y_true:y_true_batch,keep_prob:1.0}      sess.run(optimizer,feed_dict=feed_dict_train_op)      # Print status every 100 iterations.      if i%100==0:        # Calculate the accuracy on the training-set.        acc = sess.run(accuracy,feed_dict=feed_dict_train)        # Message for printing.        msg = "Optimization Iteration:{0:>6}, Training Accuracy: {1:>6.1%}"        # Print it.        print(msg.format(i+1,acc))    # Update the total number of iterations performed    total_iterations += num_iterations    # Ending time    end_time = time.time()    # Difference between start and end_times.    time_dif = end_time-start_time    # Print the time-usage    print("Time usage:"+str(timedelta(seconds=int(round(time_dif)))))  test_batch_size = 256  def print_test_accuracy():    # Number of images in the test-set.    num_test = len(data.test.images)    cls_pred = np.zeros(shape=num_test,dtype=np.int)    i = 0    while i < num_test:      # The ending index for the next batch is denoted j.      j = min(i+test_batch_size,num_test)      # Get the images from the test-set between index i and j      images = data.test.images[i:j, :]      # Get the associated labels      labels = data.test.labels[i:j, :]      # Create a feed-dict with these images and labels.      feed_dict={x:images,y_true:labels,keep_prob:1.0}      # Calculate the predicted class using Tensorflow.      cls_pred[i:j] = sess.run(y_pred_cls,feed_dict=feed_dict)      # Set the start-index for the next batch to the      # end-index of the current batch      i = j    cls_true = data.test.cls    correct = (cls_true==cls_pred)    correct_sum = correct.sum()    acc = float(correct_sum) / num_test    # Print the accuracy    msg = "Accuracy on Test-Set: {0:.1%} ({1}/{2})"    print(msg.format(acc,correct_sum,num_test))  # Performance after 10000 optimization iterations  optimize(num_iterations=10000)  print_test_accuracy()  savew_hl1 = layer_conv1["weights"].eval()  saveb_hl1 = layer_conv1["biases"].eval()  savew_hl2 = layer_conv2["weights"].eval()  saveb_hl2 = layer_conv2["biases"].eval()  savew_fc1 = fc1_layer["weights"].eval()  saveb_fc1 = fc1_layer["biases"].eval()  savew_op = fc2_layer["weights"].eval()  saveb_op = fc2_layer["biases"].eval()

 np.save("savew_hl1.npy", savew_hl1)  np.save("saveb_hl1.npy", saveb_hl1)  np.save("savew_hl2.npy", savew_hl2)  np.save("saveb_hl2.npy", saveb_hl2)  np.save("savew_hl3.npy", savew_fc1)  np.save("saveb_hl3.npy", saveb_fc1)  np.save("savew_op.npy", savew_op)  np.save("saveb_op.npy", saveb_op)