48. Fashion MNIST || convolution

2021. 11. 26. 21:03

728x90

# Fashion Mnist with cnn

import tensorflow as tf
fashion_mnist = tf.keras.datasets.fashion_mnist
(train_x, train_y), (test_x, test_y) = fashion_mnist.load_data()
train_x = train_x / 255.0
test_x = test_x / 255.0

print(train_x.shape, test_x.shape) # 2차원 이미지
train_x = train_x.reshape(-1, 28, 28, 1) # 2차원 이미지 => 3차원 이미지
test_x = test_x.reshape(-1, 28, 28, 1)
print(train_x.shape, test_x.shape)

# (60000, 28, 28) (10000, 28, 28)
# (60000, 28, 28, 1) (10000, 28, 28, 1)

# 이미지로 데이터 확인
import matplotlib.pyplot as plt
plt.figure(figsize = ( 10, 10 ))
for c in range(16) :
    plt.subplot(4,4,c+1)
    plt.imshow(train_x[c].reshape(28, 28), cmap='gray')
plt.show()

print(train_y[:16])

# [9 0 0 3 0 2 7 2 5 5 0 9 5 5 7 9]

# cnn 모델 정의
# kernel_size = 가중치 필터 3행3열

model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(input_shape=(28,28,1), kernel_size=(3,3), filters=16), # 합성곱층 3개
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=32),
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=64),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(units=128, activation='relu'),
    tf.keras.layers.Dense(units=10, activation='softmax')
])
model.compile(optimizer=tf.keras.optimizers.Adam(),
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])
model.summary()

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_10 (Conv2D)           (None, 26, 26, 16)        160       
_________________________________________________________________
conv2d_11 (Conv2D)           (None, 24, 24, 32)        4640      
_________________________________________________________________
conv2d_12 (Conv2D)           (None, 22, 22, 64)        18496     
_________________________________________________________________
flatten_1 (Flatten)          (None, 30976)             0         
_________________________________________________________________
dense (Dense)                (None, 128)               3965056   
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1290      
=================================================================
Total params: 3,989,642
Trainable params: 3,989,642
Non-trainable params: 0
_________________________________________________________________

history=model.fit(train_x, train_y, epochs=10, validation_split=0.25)

Epoch 1/10
1407/1407 [==============================] - 161s 114ms/step - loss: 0.4653 - accuracy: 0.8360 - val_loss: 0.4354 - val_accuracy: 0.8505
Epoch 2/10
1407/1407 [==============================] - 161s 114ms/step - loss: 0.3381 - accuracy: 0.8775 - val_loss: 0.4108 - val_accuracy: 0.8515
Epoch 3/10
1407/1407 [==============================] - 159s 113ms/step - loss: 0.2825 - accuracy: 0.8956 - val_loss: 0.3780 - val_accuracy: 0.8667
Epoch 4/10
1407/1407 [==============================] - 158s 112ms/step - loss: 0.2400 - accuracy: 0.9120 - val_loss: 0.4034 - val_accuracy: 0.8691
Epoch 5/10
1407/1407 [==============================] - 158s 112ms/step - loss: 0.2070 - accuracy: 0.9235 - val_loss: 0.4951 - val_accuracy: 0.8657
Epoch 6/10
1407/1407 [==============================] - 158s 112ms/step - loss: 0.1814 - accuracy: 0.9320 - val_loss: 0.4982 - val_accuracy: 0.8533
Epoch 7/10
1407/1407 [==============================] - 158s 112ms/step - loss: 0.1623 - accuracy: 0.9398 - val_loss: 0.4963 - val_accuracy: 0.8640
Epoch 8/10
1407/1407 [==============================] - 158s 112ms/step - loss: 0.1448 - accuracy: 0.9460 - val_loss: 0.5650 - val_accuracy: 0.8687
Epoch 9/10
1407/1407 [==============================] - 157s 112ms/step - loss: 0.1350 - accuracy: 0.9511 - val_loss: 0.6309 - val_accuracy: 0.8693
Epoch 10/10
1407/1407 [==============================] - 157s 112ms/step - loss: 0.1174 - accuracy: 0.9574 - val_loss: 0.6193 - val_accuracy: 0.8647

import matplotlib.pyplot as plt
plt.figure(figsize=(12,4))
plt.subplot(1,2,1)
plt.plot(history.history['loss'], 'b--', label='loss')
plt.plot(history.history['val_loss'], 'r--', label='val_loss')
plt.xlabel('Epoch')
plt.legend()

plt.subplot(1,2,2)
plt.plot(history.history['accuracy'], 'b--', label='accuracy')
plt.plot(history.history['val_accuracy'], 'r--', label='val_accuracy')
plt.xlabel('Epoch')
plt.ylim(0.7, 1)
plt.legend()
plt.show()

# cnn 성능향상방법

# 1. vggnet방식 사용 : 14개까지 깊은 신경망

# 2. 이미지 보강하기 : 하나의 이미지를 여러개의 이미지로 변형

# cnn 모델 정의
# 풀링레이어, 드랍아웃레이어 추가
model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(input_shape=(28,28,1), kernel_size=(3,3), filters=32), # 합성곱층 3개
    tf.keras.layers.MaxPool2D(strides=(2,2)), # 특성맵의 크기가 반으로 줄어듬
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=64), # 패딩이없어서 2개씩 감소
    tf.keras.layers.MaxPool2D(strides=(2,2)), # 특징맵을 2행2열로나눠서 그중 최대값만 선택한 특징맵
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=128), #

    tf.keras.layers.Flatten(), # 9x128
    tf.keras.layers.Dense(units=128, activation='relu'), 
    tf.keras.layers.Dropout(rate=0.3), # 일부러 누락
    tf.keras.layers.Dense(units=10, activation='softmax') # 출력층
])
model.compile(optimizer=tf.keras.optimizers.Adam(),
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

model.summary()


Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_3 (Conv2D)           (None, 26, 26, 32)        320       
                                                                 
 max_pooling2d (MaxPooling2D  (None, 13, 13, 32)       0         
 )                                                               
                                                                 
 conv2d_4 (Conv2D)           (None, 11, 11, 64)        18496     
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 5, 5, 64)         0         
 2D)                                                             
                                                                 
 conv2d_5 (Conv2D)           (None, 3, 3, 128)         73856     
                                                                 
 flatten_1 (Flatten)         (None, 1152)              0         
                                                                 
 dense_2 (Dense)             (None, 128)               147584    
                                                                 
 dropout (Dropout)           (None, 128)               0         
                                                                 
 dense_3 (Dense)             (None, 10)                1290      
                                                                 
=================================================================
Total params: 241,546
Trainable params: 241,546
Non-trainable params: 0
_________________________________________________________________

history=model.fit(train_x, train_y, epochs=10, validation_split=0.25)

Epoch 1/10
1407/1407 [==============================] - 48s 34ms/step - loss: 0.5216 - accuracy: 0.8115 - val_loss: 0.3656 - val_accuracy: 0.8667
Epoch 2/10
1407/1407 [==============================] - 48s 34ms/step - loss: 0.3579 - accuracy: 0.8712 - val_loss: 0.3305 - val_accuracy: 0.8783
Epoch 3/10
1407/1407 [==============================] - 48s 34ms/step - loss: 0.3150 - accuracy: 0.8869 - val_loss: 0.3104 - val_accuracy: 0.8909
Epoch 4/10
1407/1407 [==============================] - 47s 34ms/step - loss: 0.2887 - accuracy: 0.8947 - val_loss: 0.3002 - val_accuracy: 0.8916
Epoch 5/10
1407/1407 [==============================] - 47s 34ms/step - loss: 0.2686 - accuracy: 0.9012 - val_loss: 0.3242 - val_accuracy: 0.8816
Epoch 6/10
1407/1407 [==============================] - 47s 34ms/step - loss: 0.2530 - accuracy: 0.9064 - val_loss: 0.3300 - val_accuracy: 0.8905
Epoch 7/10
1407/1407 [==============================] - 47s 33ms/step - loss: 0.2344 - accuracy: 0.9127 - val_loss: 0.2962 - val_accuracy: 0.8948
Epoch 8/10
1407/1407 [==============================] - 47s 33ms/step - loss: 0.2244 - accuracy: 0.9170 - val_loss: 0.3060 - val_accuracy: 0.8957
Epoch 9/10
1407/1407 [==============================] - 47s 33ms/step - loss: 0.2116 - accuracy: 0.9216 - val_loss: 0.3231 - val_accuracy: 0.8984
Epoch 10/10
1407/1407 [==============================] - 47s 33ms/step - loss: 0.2025 - accuracy: 0.9259 - val_loss: 0.3317 - val_accuracy: 0.8967

import matplotlib.pyplot as plt
plt.figure(figsize=(12,4))
plt.subplot(1,2,1)
plt.plot(history.history['loss'], 'b--', label='loss')
plt.plot(history.history['val_loss'], 'r--', label='val_loss')
plt.xlabel('Epoch')
plt.legend()

plt.subplot(1,2,2)
plt.plot(history.history['accuracy'], 'b--', label='accuracy')
plt.plot(history.history['val_accuracy'], 'r--', label='val_accuracy')
plt.xlabel('Epoch')
plt.ylim(0.7, 1)
plt.legend()
plt.show()

model.evaluate(test_x, test_y, verbose=0)

# [0.34508538246154785, 0.89410001039505]

model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(input_shape=(28,28,1), kernel_size=(3,3), filters=32,
                          padding='same', activation='relu'), # padding=same이라서 28 안줄어즘
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=64, padding='same', activation='relu'),
    tf.keras.layers.MaxPool2D(pool_size=(2,2)), 
    tf.keras.layers.Dropout(rate=0.5),
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=128, padding='same', activation='relu'), 
    tf.keras.layers.Conv2D(kernel_size=(3,3), filters=256, padding='valid', activation='relu'), 
    tf.keras.layers.MaxPool2D(strides=(2,2)), 
    tf.keras.layers.Dropout(rate=0.5),
    tf.keras.layers.Flatten(),
    
    tf.keras.layers.Dense(units=512, activation='relu'), 
    tf.keras.layers.Dropout(rate=0.5), # 일부러 누락
    tf.keras.layers.Dense(units=256, activation='relu'), 
    tf.keras.layers.Dropout(rate=0.5), # 일부러 누락
    tf.keras.layers.Dense(units=10, activation='softmax') # 출력층
])
model.compile(optimizer=tf.keras.optimizers.Adam(),
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])
model.summary()

Model: "sequential_2"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_6 (Conv2D)           (None, 28, 28, 32)        320       
                                                                 
 conv2d_7 (Conv2D)           (None, 28, 28, 64)        18496     
                                                                 
 max_pooling2d_2 (MaxPooling  (None, 14, 14, 64)       0         
 2D)                                                             
                                                                 
 dropout_1 (Dropout)         (None, 14, 14, 64)        0         
                                                                 
 conv2d_8 (Conv2D)           (None, 14, 14, 128)       73856     
                                                                 
 conv2d_9 (Conv2D)           (None, 12, 12, 256)       295168    
                                                                 
 max_pooling2d_3 (MaxPooling  (None, 6, 6, 256)        0         
 2D)                                                             
                                                                 
 dropout_2 (Dropout)         (None, 6, 6, 256)         0         
                                                                 
 flatten_2 (Flatten)         (None, 9216)              0         
                                                                 
 dense_4 (Dense)             (None, 512)               4719104   
                                                                 
 dropout_3 (Dropout)         (None, 512)               0         
                                                                 
 dense_5 (Dense)             (None, 256)               131328    
                                                                 
 dropout_4 (Dropout)         (None, 256)               0         
                                                                 
 dense_6 (Dense)             (None, 10)                2570      
                                                                 
=================================================================
Total params: 5,240,842
Trainable params: 5,240,842
Non-trainable params: 0
_________________________________________________________________

history=model.fit(train_x, train_y, epochs=10, validation_split=0.25)

Epoch 1/10
1407/1407 [==============================] - 543s 386ms/step - loss: 0.5847 - accuracy: 0.7843 - val_loss: 0.3323 - val_accuracy: 0.8735
Epoch 2/10
1407/1407 [==============================] - 540s 384ms/step - loss: 0.3726 - accuracy: 0.8672 - val_loss: 0.2894 - val_accuracy: 0.8891
Epoch 3/10
1407/1407 [==============================] - 540s 384ms/step - loss: 0.3268 - accuracy: 0.8821 - val_loss: 0.2642 - val_accuracy: 0.9027
Epoch 4/10
1407/1407 [==============================] - 542s 386ms/step - loss: 0.3048 - accuracy: 0.8894 - val_loss: 0.2433 - val_accuracy: 0.9103
Epoch 5/10
1407/1407 [==============================] - 541s 385ms/step - loss: 0.2897 - accuracy: 0.8959 - val_loss: 0.2278 - val_accuracy: 0.9140
Epoch 6/10
1407/1407 [==============================] - 543s 386ms/step - loss: 0.2773 - accuracy: 0.9004 - val_loss: 0.2281 - val_accuracy: 0.9122
Epoch 7/10
1407/1407 [==============================] - 543s 386ms/step - loss: 0.2666 - accuracy: 0.9039 - val_loss: 0.2333 - val_accuracy: 0.9160
Epoch 8/10
1407/1407 [==============================] - 548s 389ms/step - loss: 0.2672 - accuracy: 0.9039 - val_loss: 0.2255 - val_accuracy: 0.9179
Epoch 9/10
1407/1407 [==============================] - 553s 393ms/step - loss: 0.2552 - accuracy: 0.9094 - val_loss: 0.2200 - val_accuracy: 0.9205
Epoch 10/10
1407/1407 [==============================] - 554s 394ms/step - loss: 0.2532 - accuracy: 0.9078 - val_loss: 0.2148 - val_accuracy: 0.9227

import matplotlib.pyplot as plt
plt.figure(figsize=(12,4))
plt.subplot(1,2,1)
plt.plot(history.history['loss'], 'b--', label='loss')
plt.plot(history.history['val_loss'], 'r--', label='val_loss')
plt.xlabel('Epoch')
plt.legend()

plt.subplot(1,2,2)
plt.plot(history.history['accuracy'], 'g-', label='accuracy')
plt.plot(history.history['val_accuracy'], 'k--', label='val_accuracy')
plt.xlabel('Epoch')
plt.ylim(0.7, 1)
plt.legend()
plt.show()

model.evaluate(test_x, test_y, verbose=0)

# [0.23406749963760376, 0.9172999858856201]

저작자표시 비영리

'Data_Science > Data_Analysis_Py' 카테고리의 다른 글

50. ImageDataGenerator (0)	2021.12.07
49. cifar10 \|\| convolution (0)	2021.11.26
47. Cat image, convolution 설명 (0)	2021.11.25
46. 와인품종 예측 딥러닝 이항분류 \|\| (0)	2021.11.25
45. 보스턴주택가격 딥러닝 회귀 예측 평가 (0)	2021.11.25

내 블로그 - 관리자 홈 전환	`Q` `Q`
새 글 쓰기	`W` `W`

글 수정 (권한 있는 경우)	`E` `E`
댓글 영역으로 이동	`C` `C`

이 페이지의 URL 복사	`S` `S`
맨 위로 이동	`T` `T`
티스토리 홈 이동	`H` `H`
단축키 안내	`Shift` + `/` `⇧` + `/`

My_Flow

48. Fashion MNIST || convolution

# Fashion Mnist with cnn

'Data_Science > Data_Analysis_Py' 카테고리의 다른 글

+ Recent posts

티스토리툴바

단축키

내 블로그

블로그 게시글

모든 영역