[인공지능] 모델 구성 코드 및 분석(1)

모델 구성표

입력층

Conv1

Batch Normalization

Max Pooling

Conv2

Batch Normalization

Max Pool ing

Conv3

Batch Normalization

Max Pooling

Flatten

FC

Drop out

FC

32x32

3 x32 x32

3x3x 64

3x3x128

64

0.25

10

* padding = ’valid‘, strides = (1,1), activation = 'relu’

* 최적화 함수 = Adam, 학습율 = 0.001

 

전체 코드

import tensorflow as tf
import numpy as np
from keras.models import Sequential
from keras.layers import Dense

# MNIST 데이터셋을 로드하여 준비. 샘플 값을 정수에서 부동소수로 변환

(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data()
class_name = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truk']

import matplotlib.pylab as plt

plt.figure(figsize=(6,1))
for i in range(36):
  plt.subplot(3, 12, i+1)
  plt.imshow(train_images[i], cmap="brg")
  plt.axis("off")
plt.show()

train_images = train_images.reshape((50000, 32, 32, 3))
test_images = test_images.reshape((10000, 32, 32, 3))
train_images, test_images = train_images/255.0, test_images/255.0

from keras.utils import to_categorical

# 층을 차례대로 쌓아 tf.keras.Sequential 모델을 만든다..
# 훈련에 사용할 옵티마이저(optimizer)와 손실 함수를 선택:
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(32, (3, 3), activation='relu', padding='valid', strides=(1,1), input_shape=(32, 32, 3)))
model.add(tf.keras.layers.BatchNormalization())
# model.add(tf.keras.layers.Dropout(0.25)) # BatchNormalization과 비교
model.add(tf.keras.layers.MaxPooling2D((2, 2)))

model.add(tf.keras.layers.Conv2D(64, (3, 3), activation='relu'))
model.add(tf.keras.layers.BatchNormalization())
# model.add(tf.keras.layers.Dropout(0.25)) # BatchNormalization과 비교
model.add(tf.keras.layers.MaxPooling2D((2, 2)))

model.add(tf.keras.layers.Conv2D(128, (3, 3), activation='relu'))
model.add(tf.keras.layers.BatchNormalization())
# model.add(tf.keras.layers.Dropout(0.25)) # BatchNormalization과 비교
model.add(tf.keras.layers.MaxPooling2D((2, 2)))

model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(64, activation='relu'))
model.add(tf.keras.layers.Dropout(0.25))
model.add(tf.keras.layers.Dense(10, activation='softmax'))

model.compile(optimizer=tf.optimizers.Adam(learning_rate=0.001), loss='sparse_categorical_crossentropy', metrics=['accuracy'])

model.summary()
history = model.fit(train_images, train_labels, epochs=5, batch_size=32)
plt.figure(figsize=(12,4)) # 그래프의 가로세로 비율
plt.subplot(1,1,1) # 1행1열의 첫 번째 위치
plt.plot(history.history['loss'], 'b--', label='loss') # loss는 파란색 점선
plt.plot(history.history['accuracy'], 'g-', label='Accuracy') # accuracy는 녹색실선
plt.xlabel('Epoch')
plt.legend()
plt.show()
print('최적화 완료!')

print('\n=====test result=====')
labels=model.predict(test_images)
# verbose=2로 설정하여 진행 막대(progress bar)가 나오지 않도록 설정한다.
print('\n Accuracy: %.4f' % (model.evaluate(test_images, test_labels, verbose=2)[1]))
# [0]: loss, [1]: accuracy

fig=plt.figure()
for i in range(15):
  subplot=fig.add_subplot(3, 5, i+1)
  subplot.set_xticks([])
  subplot.set_yticks([])
  subplot.set_title('%s' % class_name[np.argmax(labels[i])])
  subplot.imshow(test_images[i].reshape((32, 32, 3)), cmap=plt.cm.brg)
plt.show()

print('\n=====test result=====')
labels=model.predict(test_images)
# verbose=2로 설정하여 진행 막대(progress bar)가 나오지 않도록 설정한다.
print('\n Accuracy: %.4f' % (model.evaluate(test_images, test_labels, verbose=2)[1]))
# [0]: loss, [1]: accuracy​

 

전체 결과

Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
170498071/170498071 [==============================] - 4s 0us/step

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 30, 30, 32)        896       
                                                                 
 batch_normalization (BatchN  (None, 30, 30, 32)       128       
 ormalization)                                                   
                                                                 
 max_pooling2d (MaxPooling2D  (None, 15, 15, 32)       0         
 )                                                               
                                                                 
 conv2d_1 (Conv2D)           (None, 13, 13, 64)        18496     
                                                                 
 batch_normalization_1 (Batc  (None, 13, 13, 64)       256       
 hNormalization)                                                 
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 6, 6, 64)         0         
 2D)                                                             
                                                                 
 conv2d_2 (Conv2D)           (None, 4, 4, 128)         73856     
                                                                 
 batch_normalization_2 (Batc  (None, 4, 4, 128)        512       
 hNormalization)                                                 
                                                                 
 max_pooling2d_2 (MaxPooling  (None, 2, 2, 128)        0         
 2D)                                                             
                                                                 
 flatten (Flatten)           (None, 512)               0         
                                                                 
 dense (Dense)               (None, 64)                32832     
                                                                 
 dropout (Dropout)           (None, 64)                0         
                                                                 
 dense_1 (Dense)             (None, 10)                650       
                                                                 
=================================================================
Total params: 127,626
Trainable params: 127,178
Non-trainable params: 448
_________________________________________________________________
Epoch 1/5
1563/1563 [==============================] - 17s 4ms/step - loss: 1.4809 - accuracy: 0.4772
Epoch 2/5
1563/1563 [==============================] - 7s 4ms/step - loss: 1.1268 - accuracy: 0.6047
Epoch 3/5
1563/1563 [==============================] - 7s 4ms/step - loss: 0.9659 - accuracy: 0.6616
Epoch 4/5
1563/1563 [==============================] - 7s 4ms/step - loss: 0.8675 - accuracy: 0.7000
Epoch 5/5
1563/1563 [==============================] - 7s 4ms/step - loss: 0.7808 - accuracy: 0.7257

최적화 완료!

=====test result=====
313/313 [==============================] - 1s 2ms/step
313/313 - 1s - loss: 0.8938 - accuracy: 0.6941 - 911ms/epoch - 3ms/step

 Accuracy: 0.6941