MNIST (LeNet)
- 5 계층 구조: Conv-Pool-Conv- Pool-Conv-FC-FC(SM)
- 입력 : 32x32 필기체 숫자 영상 (MNIST 데이터)
- 풀링 : 가중치x(2x2블록의 합) + 편차항
- 시그모이드 활성화 함수 사용
- 성능: 오차율 0.95%(정확도: 99.05%)
Library Import
1
2
3
4
5
6
7
8
from tensorflow.keras import Model
from tensorflow.keras.layers import Conv2D, MaxPool2D, Flatten, Dense
from tensorflow.keras.callbacks import EarlyStopping, TensorBoard
from tensorflow.keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt
plt.style.use('seaborn')
Data Load
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
# Data Load
(X_train, y_train), (X_test, y_test) = mnist.load_data()
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)
# 모델에 이미지를 넣어줄 땐 컬러(channel=3)가 아닌 흑백 이미지(channel=1)
# 텐서 모양이 4개가 있어야 함
X_train, X_test = X_train[..., np.newaxis], X_test[..., np.newaxis]
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)
1
2
3
4
5
6
7
8
9
10
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11490434/11490434 [==============================] - 2s 0us/step
(60000, 28, 28)
(60000,)
(10000, 28, 28)
(10000,)
(60000, 28, 28, 1)
(60000,)
(10000, 28, 28, 1)
(10000,)
1
2
3
# 첫 번째 이미지
plt.imshow(X_train[0, :, :, 0])
plt.show()
Image Preprocessing
normalization : 0 ~ 255 사이 값을 0 ~ 1로 값을 변경
1
X_train, X_test = X_train / 255.0, X_test / 255.0
Model
1
2
3
num_class = 10 # 0 ~ 9까지의 10개의 class
epochs = 30
batch_size = 32
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
class LeNet5(Model):
# Model 정의
def __init__(self, num_class):
super(LeNet5, self).__init__()
self.conv1 = Conv2D(filters=6, kernel_size=(5, 5), padding='same', activation='relu')
self.conv2 = Conv2D(filters=16, kernel_size=(5, 5), activation='relu')
self.max_pool = MaxPool2D(pool_size=(2, 2))
self.flatten = Flatten()
self.dense1 = Dense(units=120, activation='relu')
self.dense2 = Dense(units=84, activation='relu')
self.dense3 = Dense(units=num_class, activation='softmax')
# Model Call
def call(self, input_data):
# Input data -> Conv2D -> MaxPool2D
x = self.max_pool(self.conv1(input_data))
# Conv2D -> MaxPool2D
x = self.max_pool(self.conv2(x))
x = self.flatten(x)
# Flatten -> Dense1 -> Dense2 -> Dense3
x = self.dense3(self.dense2(self.dense1(x)))
return x
1
model = LeNet5(num_class)
1
2
3
4
# Model Compile
model.compile(optimizer='sgd',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
1
2
3
# Callbacks
callbacks = [EarlyStopping(patience=3, monitor='val_loss'),
TensorBoard(log_dir='./logs', histogram_freq=1)]
1
2
3
4
5
6
# Model Fit
model.fit(X_train, y_train,
batch_size=batch_size,
epochs=epochs,
validation_data=(X_test, y_test),
callbacks=callbacks)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Epoch 1/30
1875/1875 [==============================] - 19s 5ms/step - loss: 0.4899 - accuracy: 0.8537 - val_loss: 0.1340 - val_accuracy: 0.9595
Epoch 2/30
1875/1875 [==============================] - 9s 5ms/step - loss: 0.1245 - accuracy: 0.9615 - val_loss: 0.1257 - val_accuracy: 0.9611
Epoch 3/30
1875/1875 [==============================] - 8s 5ms/step - loss: 0.0914 - accuracy: 0.9720 - val_loss: 0.0632 - val_accuracy: 0.9805
Epoch 4/30
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0737 - accuracy: 0.9776 - val_loss: 0.0630 - val_accuracy: 0.9795
Epoch 5/30
1875/1875 [==============================] - 8s 5ms/step - loss: 0.0634 - accuracy: 0.9804 - val_loss: 0.0517 - val_accuracy: 0.9825
Epoch 6/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0562 - accuracy: 0.9824 - val_loss: 0.0435 - val_accuracy: 0.9854
Epoch 7/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0502 - accuracy: 0.9843 - val_loss: 0.0575 - val_accuracy: 0.9813
Epoch 8/30
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0455 - accuracy: 0.9855 - val_loss: 0.0500 - val_accuracy: 0.9809
Epoch 9/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0407 - accuracy: 0.9870 - val_loss: 0.0378 - val_accuracy: 0.9872
Epoch 10/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0374 - accuracy: 0.9882 - val_loss: 0.0362 - val_accuracy: 0.9885
Epoch 11/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0343 - accuracy: 0.9892 - val_loss: 0.0405 - val_accuracy: 0.9871
Epoch 12/30
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0314 - accuracy: 0.9899 - val_loss: 0.0391 - val_accuracy: 0.9870
Epoch 13/30
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0290 - accuracy: 0.9908 - val_loss: 0.0397 - val_accuracy: 0.9871
1
%tensorboard --logdir logs
