In [1]:
import tensorflow as tf
import numpy as np
import matplotlib as plt
import matplotlib.pyplot as plt

In [2]:
%matplotlib inline
sess = tf.InteractiveSession()
image = np.array([[[[1],[2],[3]],[[4],[5],[6]],[[7],[8],[9]]]], dtype = np.float32)
print(image.shape)
plt.imshow(image.reshape(3,3), cmap='Greys') #시각화
#1 => n개의 이미지#
# 3*3 size
#1 color


(1, 3, 3, 1)
Out[2]:
<matplotlib.image.AxesImage at 0x7fde1419da90>

In [3]:
print("image.shape", image.shape)
weight = tf.constant([[[[1.]],[[1.]]],[[[1.]],[[1.]]]])
print("weight.shape", weight.shape)
conv2d = tf.nn.conv2d(image, weight, strides=[1,1,1,1],padding = 'SAME') #1*1 stride줌 #same을 했기때문에 input과 같은 사이즈나옴
conv2d_img = conv2d.eval()
print("conv2d_img.shape", conv2d_img.shape)
#시각화하기위한코드
conv2d_img = np.swapaxes(conv2d_img,0,3) #transposing 2D array
for i, one_img in enumerate(conv2d_img):
    print(one_img.reshape(3,3))
    plt.subplot(1,3,i+1), plt.imshow(one_img.reshape(3,3), cmap = 'gray')


image.shape (1, 3, 3, 1)
weight.shape (2, 2, 1, 1)
conv2d_img.shape (1, 3, 3, 1)
[[ 12.  16.   9.]
 [ 24.  28.  15.]
 [ 15.  17.   9.]]

In [4]:
print("image.shape", image.shape)
weight = tf.constant([[[[1.,10,-1.]],[[1.,10.,-1]]],[[[1.,10.,-1.]],[[1.,10.,-1.]]]])
print("weight.shape", weight.shape) #마지막 숫자 3은 필터의 갯수==나올 이미지의 갯수
#필터를 몇장을 쓰는가에 따라 하나의 이미지에 대하여 여러 이미지가 나옴.
conv2d = tf.nn.conv2d(image, weight, strides=[1,1,1,1],padding = 'SAME')
conv2d_img = conv2d.eval()
print("conv2d_img.shape", conv2d_img.shape)
#시각화 하기 위한 코드
conv2d_img = np.swapaxes(conv2d_img,0,3)
for i, one_img in enumerate(conv2d_img):
    print(one_img.reshape(3,3))
    plt.subplot(1,3,i+1), plt.imshow(one_img.reshape(3,3), cmap= 'gray')


image.shape (1, 3, 3, 1)
weight.shape (2, 2, 1, 3)
conv2d_img.shape (1, 3, 3, 3)
[[ 12.  16.   9.]
 [ 24.  28.  15.]
 [ 15.  17.   9.]]
[[ 120.  160.   90.]
 [ 240.  280.  150.]
 [ 150.  170.   90.]]
[[-12. -16.  -9.]
 [-24. -28. -15.]
 [-15. -17.  -9.]]

In [5]:
#max pooling_어떤 데이터를 서브 샘플링한다.
%matplotlib inline
image = np.array([[[[4],[3]],[[2],[1]]]],dtype = np.float32)
pool = tf.nn.max_pool(image, ksize= [1,2,2,1], strides = [1,1,1,1], padding = 'SAME') 
                 #필터사이즈 #max_pool이 CNN과 잘 동작함
print(pool.shape)
print(pool.eval())
#최댓값 뽑아내는 것


(1, 2, 2, 1)
[[[[ 4.]
   [ 3.]]

  [[ 2.]
   [ 1.]]]]

In [6]:
#max pooling2
%matplotlib inline
image = np.array([[[[4],[3]],[[2],[1]]]],dtype = np.float32)
pool = tf.nn.max_pool(image, ksize= [1,2,2,1], strides = [1,1,1,1], padding = 'VALID')
print(pool.shape)
print(pool.eval())


(1, 1, 1, 1)
[[[[ 4.]]]]

In [7]:
#실전 이미지에 넣기
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/",one_hot=True)
#데이터 읽어오기


Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz

In [8]:
img = mnist.train.images[5].reshape(28,28)
#가장 첫번째에 있는 거 불러와서 쉐잎을 28*28로 잡
plt.imshow(img, cmap = 'gray') #출력


Out[8]:
<matplotlib.image.AxesImage at 0x7fddbcd1a358>

In [9]:
sess = tf.InteractiveSession()

img = img.reshape(-1,28,28,1) #28*28의 한 색깔, n개의 이미지일 때는 -1=(컴퓨터에게 알아서 계산해~라는 reshape의 방법)
W1 = tf.Variable(tf.random_normal([3, 3, 1, 5], stddev=0.01)) #칼라에 신경써야함.(1) (3*3은 필터의 사이즈) (5개의 필터사용)
conv2d = tf.nn.conv2d(img, W1, strides=[1, 2, 2, 1], padding='SAME')# (2*2는 필터를 2칸씩 옮기겠다.그래서 출력이 14*14)
print(conv2d)
sess.run(tf.global_variables_initializer())
#그림 출력하기는 code
conv2d_img = conv2d.eval()
conv2d_img = np.swapaxes(conv2d_img, 0, 3)
for i, one_img in enumerate(conv2d_img):
    plt.subplot(1,5,i+1), plt.imshow(one_img.reshape(14,14), cmap='gray')
#5개의 서로다른 필터를 사용했기 때문에 서로다른 5개의 이미지가 나옴.


Tensor("Conv2D_2:0", shape=(1, 14, 14, 5), dtype=float32)

In [10]:
#max pooling
pool = tf.nn.max_pool(conv2d, ksize=[1, 2, 2, 1], strides=[
                        1, 2, 2, 1], padding='SAME')
#입력 이미지 14*14인데 또 필터가 두탄씩 움직이니까 7*7로 또 줄어듦.
print(pool)
#실행코드
sess.run(tf.global_variables_initializer())
pool_img = pool.eval()
#출력 code
pool_img = np.swapaxes(pool_img, 0, 3)
for i, one_img in enumerate(pool_img):
    plt.subplot(1,5,i+1), plt.imshow(one_img.reshape(7, 7), cmap='gray')

#7*7의 이미지가 나옴. 이미지가 서브 샘플링되어서 해상도가 떨어져있음. max pooling을 이용해서 간단한 서브 샘플링을 해봄.


Tensor("MaxPool_2:0", shape=(1, 7, 7, 5), dtype=float32)

In [11]:
import tensorflow as tf
import random
# import matplotlib.pyplot as plt

from tensorflow.examples.tutorials.mnist import input_data

tf.set_random_seed(777)  # reproducibility

mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) #vector표현가능하게one_hot=True
# Check out https://www.tensorflow.org/get_started/mnist/beginners for
# more information about the mnist dataset

# hyper parameters
learning_rate = 0.001
training_epochs = 15
batch_size = 100

# dropout (keep_prob) rate  0.7~0.5 on training, but should be 1 for testing
keep_prob = tf.placeholder(tf.float32)

#Conv layer1
#input의 이미지를 우리가 원하는대로 만들어야
# input place holders
X = tf.placeholder(tf.float32, [None, 784]) #784인 이유는 MNIST가 28*28이기 때문
X_img = tf.reshape(X, [-1, 28, 28, 1])   # img 28x28x1(칼라) (black/white) #이미지로 넣기 위해서 #x_img가 입력이 될 것.
Y = tf.placeholder(tf.float32, [None, 10])

#첫번째 conv layer1
# L1 ImgIn shape=(?, 28, 28, 1)
W1 = tf.Variable(tf.random_normal([3, 3, 1, 32], stddev=0.01)) #3*3size의 필터 #1은 칼라 #32 필터
#    Conv     -> (?, 28, 28, 32)
#    Pool     -> (?, 14, 14, 32)
L1 = tf.nn.conv2d(X_img, W1, strides=[1, 1, 1, 1], padding='SAME')
L1 = tf.nn.relu(L1)
L1 = tf.nn.max_pool(L1, ksize=[1, 2, 2, 1],
                    strides=[1, 2, 2, 1], padding='SAME')
#필터 2*2 # stride 2*2 # 28*28->14*14

L1 = tf.nn.dropout(L1, keep_prob=keep_prob)
'''
Tensor("Conv2D:0", shape=(?, 28, 28, 32), dtype=float32)
Tensor("Relu:0", shape=(?, 28, 28, 32), dtype=float32)
Tensor("MaxPool:0", shape=(?, 14, 14, 32), dtype=float32)
Tensor("dropout/mul:0", shape=(?, 14, 14, 32), dtype=float32)
'''
#conv layer2
# L2 ImgIn shape=(?, 14, 14, 32)
W2 = tf.Variable(tf.random_normal([3, 3, 32, 64], stddev=0.01)) #32는 필터의 갯수와 같아야함. 64개의 필터를 쓸 것
#    Conv      ->(?, 14, 14, 64)
#    Pool      ->(?, 7, 7, 64)
L2 = tf.nn.conv2d(L1, W2, strides=[1, 1, 1, 1], padding='SAME')
L2 = tf.nn.relu(L2)
L2 = tf.nn.max_pool(L2, ksize=[1, 2, 2, 1],
                    strides=[1, 2, 2, 1], padding='SAME')
L2 = tf.nn.dropout(L2, keep_prob=keep_prob)
'''
Tensor("Conv2D_1:0", shape=(?, 14, 14, 64), dtype=float32)
Tensor("Relu_1:0", shape=(?, 14, 14, 64), dtype=float32)
Tensor("MaxPool_1:0", shape=(?, 7, 7, 64), dtype=float32)
Tensor("dropout_1/mul:0", shape=(?, 7, 7, 64), dtype=float32)
'''

#conv layer3
# L3 ImgIn shape=(?, 7, 7, 64)
W3 = tf.Variable(tf.random_normal([3, 3, 64, 128], stddev=0.01))
#    Conv      ->(?, 7, 7, 128)
#    Pool      ->(?, 4, 4, 128)
#    Reshape   ->(?, 4 * 4 * 128) # Flatten them for FC
L3 = tf.nn.conv2d(L2, W3, strides=[1, 1, 1, 1], padding='SAME')
L3 = tf.nn.relu(L3)
L3 = tf.nn.max_pool(L3, ksize=[1, 2, 2, 1], strides=[
                    1, 2, 2, 1], padding='SAME')
L3 = tf.nn.dropout(L3, keep_prob=keep_prob)

#입체적인 모양을 쭉 펼쳐야됨. 128*4*4만크므이 길이를 갖는게 n개 있게됨.
L3_flat = tf.reshape(L3, [-1, 128 * 4 * 4])
'''
Tensor("Conv2D_2:0", shape=(?, 7, 7, 128), dtype=float32)
Tensor("Relu_2:0", shape=(?, 7, 7, 128), dtype=float32)
Tensor("MaxPool_2:0", shape=(?, 4, 4, 128), dtype=float32)
Tensor("dropout_2/mul:0", shape=(?, 4, 4, 128), dtype=float32)
Tensor("Reshape_1:0", shape=(?, 2048), dtype=float32)
'''
#FC를 2번하겠다.그럼 정확도가 높아짐.
#Conv layer4
# L4 FC 4x4x128 inputs -> 625 outputs
#벡터 입력의 값 = 128*4*4 출력의 값 625개
W4 = tf.get_variable("W4", shape=[128 * 4 * 4, 625],
                     initializer=tf.contrib.layers.xavier_initializer())

#bias를 출력의 값과 똑같게 줌 (625)
b4 = tf.Variable(tf.random_normal([625]))
#곱하고 더한다.
L4 = tf.nn.relu(tf.matmul(L3_flat, W4) + b4)
#학습할때는 dropout을 0.5나 0.7로
L4 = tf.nn.dropout(L4, keep_prob=keep_prob)
'''
Tensor("Relu_3:0", shape=(?, 625), dtype=float32)
Tensor("dropout_3/mul:0", shape=(?, 625), dtype=float32)
'''
#Fully connecter(fc,dense) layer
# L5 Final FC 625 inputs 입력받아서 -> 10 outputs
W5 = tf.get_variable("W5", shape=[625, 10],
                     initializer=tf.contrib.layers.xavier_initializer())
b5 = tf.Variable(tf.random_normal([10]))
logits = tf.matmul(L4, W5) + b5
'''
Tensor("add_1:0", shape=(?, 10), dtype=float32)
'''

# define cost/loss & optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
    logits=logits, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

#Training and Evaluation
# initialize
sess = tf.Session()
sess.run(tf.global_variables_initializer())

# 학습시키기
print('Learning started. It takes sometime.')
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)
        feed_dict = {X: batch_xs, Y: batch_ys, keep_prob: 0.7}
        c, _ = sess.run([cost, optimizer], feed_dict=feed_dict)
        avg_cost += c / total_batch

    print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.9f}'.format(avg_cost))

print('Learning Finished!')

# 학습이 잘 되었는지 평가하기

# if you have a OOM error, please refer to lab-11-X-mnist_deep_cnn_low_memory.py

correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print('Accuracy:', sess.run(accuracy, feed_dict={
      X: mnist.test.images, Y: mnist.test.labels, keep_prob: 1}))
#test할 때는 반드시 dropout을 1로

# Get one and predict
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(logits, 1), feed_dict={X: mnist.test.images[r:r + 1], keep_prob: 1}))

# plt.imshow(mnist.test.images[r:r + 1].
#           reshape(28, 28), cmap='Greys', interpolation='nearest')
# plt.show(
#cost가 떨어지면서99프로까지 정확성을 얻을 수 있음.


Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
Learning started. It takes sometime.
Epoch: 0001 cost = 0.478379004
Epoch: 0002 cost = 0.107837441
Epoch: 0003 cost = 0.078062812
Epoch: 0004 cost = 0.064202867
Epoch: 0005 cost = 0.053259693
Epoch: 0006 cost = 0.049641101
Epoch: 0007 cost = 0.043886360
Epoch: 0008 cost = 0.039404330
Epoch: 0009 cost = 0.037726444
Epoch: 0010 cost = 0.035387208
Epoch: 0011 cost = 0.032274738
Epoch: 0012 cost = 0.033030399
Epoch: 0013 cost = 0.029083194
Epoch: 0014 cost = 0.027826021
Epoch: 0015 cost = 0.027642632
Learning Finished!
Accuracy: 0.9939
Label:  [5]
Prediction:  [5]

In [21]:
#위의 코드는 관리하기 불편함.그래서 python의 class으로 보다 효과적으로 관리.
# Lab 11 MNIST and Deep learning CNN

import tensorflow as tf
# import matplotlib.pyplot as plt

from tensorflow.examples.tutorials.mnist import input_data

tf.set_random_seed(777)  # reproducibility

mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
# Check out https://www.tensorflow.org/get_started/mnist/beginners for
# more information about the mnist dataset

# hyper parameters
learning_rate = 0.001
training_epochs = 15
batch_size = 100


class Model:

    #초기화
    def __init__(self, sess, name):
        self.sess = sess
        self.name = name
        self._build_net()
#네트워크를 빌드하는 건 다 넣음.
        
    def _build_net(self):
        with tf.variable_scope(self.name):
            # dropout (keep_prob) rate  0.7~0.5 on training, but should be 1
            # for testing
            self.keep_prob = tf.placeholder(tf.float32)

            # input place holders
            self.X = tf.placeholder(tf.float32, [None, 784])
            # img 28x28x1 (black/white)
            X_img = tf.reshape(self.X, [-1, 28, 28, 1])
            self.Y = tf.placeholder(tf.float32, [None, 10])

            # L1 ImgIn shape=(?, 28, 28, 1)
            W1 = tf.Variable(tf.random_normal([3, 3, 1, 32], stddev=0.01))
            #    Conv     -> (?, 28, 28, 32)
            #    Pool     -> (?, 14, 14, 32)
            L1 = tf.nn.conv2d(X_img, W1, strides=[1, 1, 1, 1], padding='SAME')
            L1 = tf.nn.relu(L1)
            L1 = tf.nn.max_pool(L1, ksize=[1, 2, 2, 1],
                                strides=[1, 2, 2, 1], padding='SAME')
            L1 = tf.nn.dropout(L1, keep_prob=self.keep_prob)
            '''
            Tensor("Conv2D:0", shape=(?, 28, 28, 32), dtype=float32)
            Tensor("Relu:0", shape=(?, 28, 28, 32), dtype=float32)
            Tensor("MaxPool:0", shape=(?, 14, 14, 32), dtype=float32)
            Tensor("dropout/mul:0", shape=(?, 14, 14, 32), dtype=float32)
            '''

            # L2 ImgIn shape=(?, 14, 14, 32)
            W2 = tf.Variable(tf.random_normal([3, 3, 32, 64], stddev=0.01))
            #    Conv      ->(?, 14, 14, 64)
            #    Pool      ->(?, 7, 7, 64)
            L2 = tf.nn.conv2d(L1, W2, strides=[1, 1, 1, 1], padding='SAME')
            L2 = tf.nn.relu(L2)
            L2 = tf.nn.max_pool(L2, ksize=[1, 2, 2, 1],
                                strides=[1, 2, 2, 1], padding='SAME')
            L2 = tf.nn.dropout(L2, keep_prob=self.keep_prob)
            '''
            Tensor("Conv2D_1:0", shape=(?, 14, 14, 64), dtype=float32)
            Tensor("Relu_1:0", shape=(?, 14, 14, 64), dtype=float32)
            Tensor("MaxPool_1:0", shape=(?, 7, 7, 64), dtype=float32)
            Tensor("dropout_1/mul:0", shape=(?, 7, 7, 64), dtype=float32)
            '''

            # L3 ImgIn shape=(?, 7, 7, 64)
            W3 = tf.Variable(tf.random_normal([3, 3, 64, 128], stddev=0.01))
            #    Conv      ->(?, 7, 7, 128)
            #    Pool      ->(?, 4, 4, 128)
            #    Reshape   ->(?, 4 * 4 * 128) # Flatten them for FC
            L3 = tf.nn.conv2d(L2, W3, strides=[1, 1, 1, 1], padding='SAME')
            L3 = tf.nn.relu(L3)
            L3 = tf.nn.max_pool(L3, ksize=[1, 2, 2, 1], strides=[
                                1, 2, 2, 1], padding='SAME')
            L3 = tf.nn.dropout(L3, keep_prob=self.keep_prob)

            L3_flat = tf.reshape(L3, [-1, 128 * 4 * 4])
            '''
            Tensor("Conv2D_2:0", shape=(?, 7, 7, 128), dtype=float32)
            Tensor("Relu_2:0", shape=(?, 7, 7, 128), dtype=float32)
            Tensor("MaxPool_2:0", shape=(?, 4, 4, 128), dtype=float32)
            Tensor("dropout_2/mul:0", shape=(?, 4, 4, 128), dtype=float32)
            Tensor("Reshape_1:0", shape=(?, 2048), dtype=float32)
            '''

            # L4 FC 4x4x128 inputs -> 625 outputs
            W4 = tf.get_variable("W4", shape=[128 * 4 * 4, 625],
                                 initializer=tf.contrib.layers.xavier_initializer())
            b4 = tf.Variable(tf.random_normal([625]))
            L4 = tf.nn.relu(tf.matmul(L3_flat, W4) + b4)
            L4 = tf.nn.dropout(L4, keep_prob=self.keep_prob)
            '''
            Tensor("Relu_3:0", shape=(?, 625), dtype=float32)
            Tensor("dropout_3/mul:0", shape=(?, 625), dtype=float32)
            '''

            # L5 Final FC 625 inputs -> 10 outputs
            W5 = tf.get_variable("W5", shape=[625, 10],
                                 initializer=tf.contrib.layers.xavier_initializer())
            b5 = tf.Variable(tf.random_normal([10]))
            self.logits = tf.matmul(L4, W5) + b5
            '''
            Tensor("add_1:0", shape=(?, 10), dtype=float32)
            '''

        # define cost/loss & optimizer
        self.cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
            logits=self.logits, labels=self.Y))
        self.optimizer = tf.train.AdamOptimizer(
            learning_rate=learning_rate).minimize(self.cost)

        correct_prediction = tf.equal(
            tf.argmax(self.logits, 1), tf.argmax(self.Y, 1))
        self.accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
#예측하는거
    def predict(self, x_test, keep_prop=1.0):
        return self.sess.run(self.logits, feed_dict={self.X: x_test, self.keep_prob: keep_prop})
#정확도를 얻는 것
    def get_accuracy(self, x_test, y_test, keep_prop=1.0):
        return self.sess.run(self.accuracy, feed_dict={self.X: x_test, self.Y: y_test, self.keep_prob: keep_prop})
#학습하는 서
    def train(self, x_data, y_data, keep_prop=0.7):
        return self.sess.run([self.cost, self.optimizer], feed_dict={
            self.X: x_data, self.Y: y_data, self.keep_prob: keep_prop})

# initialize
sess = tf.Session()
                                               #model1을 만듦.
m1 = Model(sess, "m1")

sess.run(tf.global_variables_initializer())

print('Learning Started!')

# train my model
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)
                                               #session run할 필요없이 ml에 train 함수를 호출해버리면 됨.
                                               #깔끔하게 관리가 됨.
        c, _ = m1.train(batch_xs, batch_ys)
        avg_cost += c / total_batch

    print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.9f}'.format(avg_cost))

print('Learning Finished!')

# Test model and check accuracy
print('Accuracy:', m1.get_accuracy(mnist.test.images, mnist.test.labels))


Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
Learning Started!
Epoch: 0001 cost = 0.404052137
Epoch: 0002 cost = 0.090525415
Epoch: 0003 cost = 0.068485891
Epoch: 0004 cost = 0.058093703
Epoch: 0005 cost = 0.050667376
Epoch: 0006 cost = 0.042372070
Epoch: 0007 cost = 0.040836713
Epoch: 0008 cost = 0.038622323
Epoch: 0009 cost = 0.035570359
Epoch: 0010 cost = 0.032131624
Epoch: 0011 cost = 0.031765728
Epoch: 0012 cost = 0.032215357
Epoch: 0013 cost = 0.027459460
Epoch: 0014 cost = 0.027606058
Epoch: 0015 cost = 0.026013258
Learning Finished!
Accuracy: 0.9951