CNN(2) - ex.1

 본 게시글은 필자가 강의, 책 등을 통해 개인적으로 학습한 것으로

본문의 모든 정보는 출처를 기반으로 작성되었습니다.

 

이제 CNN을 실습해보자.

data set으로는 MNIST를 사용한다.

 

전체 코드

import numpy as np
import tensorflow as tf
import random

mnist = tf.keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_test = x_test / 255
x_train = x_train / 255
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)

# one hot encode y data
y_train = tf.keras.utils.to_categorical(y_train, 10)
y_test = tf.keras.utils.to_categorical(y_test, 10)

# hyper parameters
learning_rate = 0.001
training_epochs = 12
batch_size = 128

tf.model = tf.keras.Sequential()
# L1
tf.model.add(tf.keras.layers.Conv2D(filters=16, kernel_size=(3, 3), input_shape=(28, 28, 1), activation='relu'))
tf.model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))

# L2
tf.model.add(tf.keras.layers.Conv2D(filters=32, kernel_size=(3, 3), activation='relu'))
tf.model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))

# L3 fully connected
tf.model.add(tf.keras.layers.Flatten())
tf.model.add(tf.keras.layers.Dense(units=10, kernel_initializer='glorot_normal', activation='softmax'))

tf.model.compile(loss='categorical_crossentropy', optimizer=tf.keras.optimizers.Adam(lr=learning_rate), metrics=['accuracy'])
tf.model.summary()

tf.model.fit(x_train, y_train, batch_size=batch_size, epochs=training_epochs)

# predict 10 random hand-writing data
y_predicted = tf.model.predict(x_test)
for x in range(0, 10):
    random_index = random.randint(0, x_test.shape[0]-1)
    print("index: ", random_index,
          "actual y: ", np.argmax(y_test[random_index]),
          "predicted y: ", np.argmax(y_predicted[random_index]))

evaluation = tf.model.evaluate(x_test, y_test)
print('loss: ', evaluation[0])
print('accuracy', evaluation[1])

 

 

 

 

몇가지만 집고 넘아가자.

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_test = x_test / 255
x_train = x_train / 255
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)

Data preprocessing 과정에서 training data를 convolutional Layer의 input에 알맞게

reshape해주었다.

 

tf.model = tf.keras.Sequential()
# L1
tf.model.add(tf.keras.layers.Conv2D(filters=16, kernel_size=(3, 3), input_shape=(28, 28, 1), activation='relu'))
tf.model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))

# L2
tf.model.add(tf.keras.layers.Conv2D(filters=32, kernel_size=(3, 3), activation='relu'))
tf.model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))

# L3 fully connected
tf.model.add(tf.keras.layers.Flatten())
tf.model.add(tf.keras.layers.Dense(units=10, kernel_initializer='glorot_normal', activation='softmax'))

Data set → Convolutional Layer → Pooling Layer → Convolutional Layer → Pooling Layer

→ Flatten Layer → FC Layer  단계로 층을 쌓아준 모습이다.

 

이하 설명은 생략한다.

 

 

 

참고 문헌 및 자료

1. Sung Kim Youtube channel : https://www.youtube.com/channel/UCML9R2ol-l0Ab9OXoNnr7Lw

Sung Kim

2. Andrew Ng Coursera class : https://www.coursera.org/learn/machine-learning

3. 조태호(2017). 모두의 딥러닝. 서울: 길벗