Lab07-2. MNIST data

MNIST data

1. MNIST Dataset

1) MNIST Data

• MNIST(Mixed National Institute of Standards and Technology database)

• 우체국에서 수기로 작성한 우편번호(숫자: 0~9)를 인식하기 위한 data set

   

 

2) MNIST Data set 가져오기

• Tensorflow에서는 MNIST에 Data set 가져올 수 있도록 라이브러리화 되어 있음

• 실행 시 지정한 경로로 data set을 다운로드 함

  import tensorflow as tf
  from tensorflow.examples.tutorials.mnist import input_data
  
  mnist = input_data.read_data_sets("Data/MNIST_data/", one_hot=True)
  

 

• Data set

데이터 종류	데이터 셋 명	차원(Shape)	파일
training set images	mnist.train.images.shape	(55000, 784)	train-images-idx3-ubyte.gz (9912422 bytes)
training set labels	mnist.train.labels.shape	(55000, 10)	train-labels-idx1-ubyte.gz (28881 bytes)
test set images	mnist.test.images.shape	(10000, 784)	t10k-images-idx3-ubyte.gz (1648877 bytes)
test set labels	mnist.test.labels.shape	(10000, 10)	t10k-labels-idx1-ubyte.gz (4542 bytes)

3) 소스코드 구현을 위한 용어정리

• One epoch: 전체 training data set을 한번 씩 모두 학습 시키는 것

• Batch size: one epoch가 크고 메모리가 한정적이므로 나눠서 학습하기 위한 단위

  • Batch size가 클수록 더 많은 메모리 공간을 필요로 함

• Iteration의 수: 전체 training data set / batch size

   

2. Example Code

1) 코드 실행 시 발생한 에러와 해결방법

• 에러메시지

URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed (_ssl.c:749)>

 

• 해결방법

  

  
  

2) 소스코드와 실행

• 소스코드

  • softmax 함수를 이용

  import tensorflow as tf
  from tensorflow.examples.tutorials.mnist import input_data
  import matplotlib.pyplot as plt
  import random
  
  mnist = input_data.read_data_sets("Data/MNIST_data/", one_hot=True)
  
  nb_classes = 10
  
  # MNIST data image of shape 28*28 = 784
  X = tf.placeholder(tf.float32, [None, 784])
  # 0-9 digits recognition = 10 classes
  Y = tf.placeholder(tf.float32, [None, nb_classes])
  
  W = tf.Variable(tf.random_normal([784, nb_classes]))
  b = tf.Variable(tf.random_normal([nb_classes]))
  
  # Hypothesis (using softmax)
  hypothesis = tf.nn.softmax(tf.matmul(X, W) + b)
  cost = tf.reduce_mean(-tf.reduce_sum(Y * tf.log(hypothesis), axis=1))
  optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(cost)
  
  # Test model
  is_correct = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
  accuracy = tf.reduce_mean(tf.cast(is_correct, tf.float32))
  
  training_epochs = 15
  batch_size = 100
  
  with tf.Session() as sess:
      sess.run(tf.global_variables_initializer())
  
      # Training cycle
      for epoch in range(training_epochs):
          avg_cost = 0
          total_batch = int(mnist.train.num_examples / batch_size)
  
          for i in range(total_batch):
              batch_xs, batch_ys = mnist.train.next_batch(batch_size)
              c, _ = sess.run([cost, optimizer], feed_dict={X: batch_xs, Y: batch_ys})
  
              avg_cost += c / total_batch
  
          print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.9f}'.format(avg_cost))
  
      # Test the model using test sets
      print("Accuracy: ", accuracy.eval(session=sess,
                                        feed_dict={X: mnist.test.images, Y: mnist.test.labels}))
  
      # sample image show and prediction
      r = random.randint(0, mnist.test.num_examples - 1)
      print("Label:", sess.run(tf.argmax(mnist.test.labels[r:r+1], 1)))
      print("Prediction:", sess.run(tf.argmax(hypothesis, 1),
                                    feed_dict={X: mnist.test.images[r:r+1]}))
  
      print("sample image shape:", mnist.test.images[r:r+1].shape)
      plt.imshow(mnist.test.images[r:r+1].reshape(28, 28), cmap='Greys', interpolation='nearest')
      plt.show()
  

   

• 결과

  Extracting Data/MNIST_data/train-images-idx3-ubyte.gz
  Extracting Data/MNIST_data/train-labels-idx1-ubyte.gz
  Extracting Data/MNIST_data/t10k-images-idx3-ubyte.gz
  Extracting Data/MNIST_data/t10k-labels-idx1-ubyte.gz
  Epoch: 0001 cost = 2.863736535
  Epoch: 0002 cost = 1.122258349
  Epoch: 0003 cost = 0.890512572
  Epoch: 0004 cost = 0.776820077
  Epoch: 0005 cost = 0.705135050
  Epoch: 0006 cost = 0.655167040
  Epoch: 0007 cost = 0.617489539
  Epoch: 0008 cost = 0.587442237
  Epoch: 0009 cost = 0.562598704
  Epoch: 0010 cost = 0.542466953
  Epoch: 0011 cost = 0.525049096
  Epoch: 0012 cost = 0.509709905
  Epoch: 0013 cost = 0.496286628
  Epoch: 0014 cost = 0.484218249
  Epoch: 0015 cost = 0.473724156
  Accuracy:  0.8853
  Label: [8]
  Prediction: [8]
  sample image shape: (1, 784)