CNN Sequential 모델 학습

In [0]:
# 런타임 -> 런타임 유형변경 -> 하드웨어 가속도 TPU변경
%tensorflow_version 2.x
#런타임 -> 런타임 다시시작
TensorFlow 2.x selected.

1. Importing Libraries

In [0]:
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
In [0]:
import tensorflow as tf 
from tensorflow import keras
from tensorflow.keras.utils import to_categorical # one-hot 인코딩
import numpy as np
import matplotlib.pyplot as plt
import os

print(tf.__version__)     # 텐서플로우 버전확인 (colab의 기본버전은 1.15.0) --> 2.0 변경 "%tensorflow_version 2.x"
print(keras.__version__)  # 케라스 버전확인
2.1.0
2.2.4-tf

2. Hyper Parameters

In [0]:
learning_rate = 0.001
training_epochs = 50
batch_size = 100

3. MNIST Data

In [0]:
mnist = keras.datasets.mnist
class_names = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
In [0]:
# MNIST image load (trian, test)
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()    

# 0~255 중 하나로 표현되는 입력 이미지들의 값을 1 이하가 되도록 정규화    
train_images = train_images.astype(np.float32) / 255.
test_images = test_images.astype(np.float32) / 255.

# np.expand_dims 차원을 변경
train_images = np.expand_dims(train_images, axis=-1)
test_images = np.expand_dims(test_images, axis=-1)

# label을 ont-hot encoding    
train_labels = to_categorical(train_labels, 10)
test_labels = to_categorical(test_labels, 10) 
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11493376/11490434 [==============================] - 0s 0us/step

4. Model Function

In [0]:
# Sequential 모델 층 구성하기
def create_model():
    model = keras.Sequential() # Sequential 모델 시작
    model.add(keras.layers.Conv2D(filters=32, kernel_size=3, activation=tf.nn.relu, padding='SAME', 
                                  input_shape=(28, 28, 1)))
    model.add(keras.layers.MaxPool2D(padding='SAME'))
    model.add(keras.layers.Conv2D(filters=64, kernel_size=3, activation=tf.nn.relu, padding='SAME'))
    model.add(keras.layers.MaxPool2D(padding='SAME'))
    model.add(keras.layers.Conv2D(filters=128, kernel_size=3, activation=tf.nn.relu, padding='SAME'))
    model.add(keras.layers.MaxPool2D(padding='SAME'))
    model.add(keras.layers.Flatten())
    model.add(keras.layers.Dense(256, activation=tf.nn.relu))
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(10, activation=tf.nn.softmax))
    return model
In [0]:
model = create_model() # 모델 함수를 model로 변경
model.summary() # 모델에 대한 요약 출력해줌
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 28, 28, 32)        320       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 14, 14, 64)        18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 7, 7, 64)          0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 7, 7, 128)         73856     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 4, 4, 128)         0         
_________________________________________________________________
flatten (Flatten)            (None, 2048)              0         
_________________________________________________________________
dense (Dense)                (None, 256)               524544    
_________________________________________________________________
dropout (Dropout)            (None, 256)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 10)                2570      
=================================================================
Total params: 619,786
Trainable params: 619,786
Non-trainable params: 0
_________________________________________________________________
In [0]:
# 위에서 정한 모델을 그림으로(plot) 보여줌
keras.utils.plot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True) 
Out[0]:

5. Training

In [0]:
# CNN 모델 구조 확정하고 컴파일 진행
model.compile(loss='categorical_crossentropy',       # crossentropy loss
              optimizer='adam',                      # adam optimizer
              metrics=['accuracy'])                  # 측정값 : accuracy

# 학습실행
history = model.fit(train_images, train_labels,       # 입력값
          batch_size=batch_size,                      # 1회마다 배치마다 100개 프로세스 
          epochs=training_epochs,                     # 50회 학습
          verbose=1,                                  # verbose는 학습 중 출력되는 문구를 설정하는 것 
          validation_data=(test_images, test_labels)) # test를 val로 사용
          
# test 값 결과 확인
score = model.evaluate(test_images, test_labels, verbose=0) # verbose가 0 이면 ==== 움직이지 않고, 1이면 ==== 진행 바가 움직임
print('Test loss:', score[0])
print('Test accuracy:', score[1])
Train on 60000 samples, validate on 10000 samples
Epoch 1/50
60000/60000 [==============================] - 93s 2ms/sample - loss: 0.1936 - accuracy: 0.9391 - val_loss: 0.0403 - val_accuracy: 0.9870
Epoch 2/50
60000/60000 [==============================] - 91s 2ms/sample - loss: 0.0557 - accuracy: 0.9836 - val_loss: 0.0286 - val_accuracy: 0.9909
Epoch 3/50
60000/60000 [==============================] - 91s 2ms/sample - loss: 0.0377 - accuracy: 0.9884 - val_loss: 0.0248 - val_accuracy: 0.9920
Epoch 4/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0301 - accuracy: 0.9906 - val_loss: 0.0249 - val_accuracy: 0.9923
Epoch 5/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0238 - accuracy: 0.9923 - val_loss: 0.0231 - val_accuracy: 0.9930
Epoch 6/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0197 - accuracy: 0.9938 - val_loss: 0.0194 - val_accuracy: 0.9943
Epoch 7/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0165 - accuracy: 0.9947 - val_loss: 0.0201 - val_accuracy: 0.9944
Epoch 8/50
60000/60000 [==============================] - 91s 2ms/sample - loss: 0.0120 - accuracy: 0.9960 - val_loss: 0.0334 - val_accuracy: 0.9912
Epoch 9/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0160 - accuracy: 0.9948 - val_loss: 0.0201 - val_accuracy: 0.9934
Epoch 10/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0096 - accuracy: 0.9973 - val_loss: 0.0214 - val_accuracy: 0.9942
Epoch 11/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0108 - accuracy: 0.9965 - val_loss: 0.0240 - val_accuracy: 0.9939
Epoch 12/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0096 - accuracy: 0.9967 - val_loss: 0.0274 - val_accuracy: 0.9938
Epoch 13/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0085 - accuracy: 0.9973 - val_loss: 0.0242 - val_accuracy: 0.9934
Epoch 14/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0069 - accuracy: 0.9975 - val_loss: 0.0333 - val_accuracy: 0.9925
Epoch 15/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0075 - accuracy: 0.9975 - val_loss: 0.0308 - val_accuracy: 0.9920
Epoch 16/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0077 - accuracy: 0.9975 - val_loss: 0.0309 - val_accuracy: 0.9927
Epoch 17/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0058 - accuracy: 0.9980 - val_loss: 0.0298 - val_accuracy: 0.9935
Epoch 18/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0061 - accuracy: 0.9981 - val_loss: 0.0240 - val_accuracy: 0.9940
Epoch 19/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0075 - accuracy: 0.9976 - val_loss: 0.0243 - val_accuracy: 0.9940
Epoch 20/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0057 - accuracy: 0.9983 - val_loss: 0.0284 - val_accuracy: 0.9942
Epoch 21/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0055 - accuracy: 0.9983 - val_loss: 0.0275 - val_accuracy: 0.9940
Epoch 22/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0046 - accuracy: 0.9986 - val_loss: 0.0286 - val_accuracy: 0.9943
Epoch 23/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0040 - accuracy: 0.9987 - val_loss: 0.0257 - val_accuracy: 0.9942
Epoch 24/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0048 - accuracy: 0.9983 - val_loss: 0.0339 - val_accuracy: 0.9924
Epoch 25/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0058 - accuracy: 0.9981 - val_loss: 0.0271 - val_accuracy: 0.9941
Epoch 26/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0036 - accuracy: 0.9988 - val_loss: 0.0353 - val_accuracy: 0.9921
Epoch 27/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0033 - accuracy: 0.9989 - val_loss: 0.0330 - val_accuracy: 0.9946
Epoch 28/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0049 - accuracy: 0.9985 - val_loss: 0.0462 - val_accuracy: 0.9915
Epoch 29/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0045 - accuracy: 0.9985 - val_loss: 0.0426 - val_accuracy: 0.9929
Epoch 30/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0039 - accuracy: 0.9987 - val_loss: 0.0298 - val_accuracy: 0.9938
Epoch 31/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0033 - accuracy: 0.9989 - val_loss: 0.0374 - val_accuracy: 0.9938
Epoch 32/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0044 - accuracy: 0.9985 - val_loss: 0.0459 - val_accuracy: 0.9919
Epoch 33/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0030 - accuracy: 0.9991 - val_loss: 0.0407 - val_accuracy: 0.9927
Epoch 34/50
60000/60000 [==============================] - 89s 1ms/sample - loss: 0.0025 - accuracy: 0.9992 - val_loss: 0.0428 - val_accuracy: 0.9927
Epoch 35/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0036 - accuracy: 0.9989 - val_loss: 0.0419 - val_accuracy: 0.9936
Epoch 36/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0045 - accuracy: 0.9988 - val_loss: 0.0351 - val_accuracy: 0.9942
Epoch 37/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0036 - accuracy: 0.9988 - val_loss: 0.0463 - val_accuracy: 0.9918
Epoch 38/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0040 - accuracy: 0.9989 - val_loss: 0.0455 - val_accuracy: 0.9931
Epoch 39/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0029 - accuracy: 0.9991 - val_loss: 0.0439 - val_accuracy: 0.9923
Epoch 40/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0030 - accuracy: 0.9991 - val_loss: 0.0399 - val_accuracy: 0.9932
Epoch 41/50
60000/60000 [==============================] - 88s 1ms/sample - loss: 0.0032 - accuracy: 0.9991 - val_loss: 0.0481 - val_accuracy: 0.9931
Epoch 42/50
60000/60000 [==============================] - 92s 2ms/sample - loss: 0.0034 - accuracy: 0.9990 - val_loss: 0.0443 - val_accuracy: 0.9926
Epoch 43/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0028 - accuracy: 0.9991 - val_loss: 0.0618 - val_accuracy: 0.9904
Epoch 44/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0054 - accuracy: 0.9985 - val_loss: 0.0352 - val_accuracy: 0.9938
Epoch 45/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0017 - accuracy: 0.9996 - val_loss: 0.0484 - val_accuracy: 0.9938
Epoch 46/50
60000/60000 [==============================] - 91s 2ms/sample - loss: 0.0035 - accuracy: 0.9991 - val_loss: 0.0443 - val_accuracy: 0.9926
Epoch 47/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0023 - accuracy: 0.9992 - val_loss: 0.0389 - val_accuracy: 0.9940
Epoch 48/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0034 - accuracy: 0.9993 - val_loss: 0.0408 - val_accuracy: 0.9937
Epoch 49/50
60000/60000 [==============================] - 90s 1ms/sample - loss: 0.0037 - accuracy: 0.9989 - val_loss: 0.0449 - val_accuracy: 0.9938
Epoch 50/50
60000/60000 [==============================] - 90s 2ms/sample - loss: 0.0038 - accuracy: 0.9989 - val_loss: 0.0463 - val_accuracy: 0.9935
Test loss: 0.046290978360095054
Test accuracy: 0.9935

6. Visualization

In [0]:
import matplotlib.pyplot as plt
import numpy as np
import os

# 모델 학습 후 정보가 담긴 history 내용을 토대로 선 그래프를 그리는 함수 설정

def plot_acc(history, title=None):        # Accuracy(정확도) Visualization
    # summarize history for accuracy
    if not isinstance(history, dict):
        history = history.history

    plt.plot(history['accuracy'])        # accuracy
    plt.plot(history['val_accuracy'])    # validation accuracy
    if title is not None:
        plt.title(title)
    plt.ylabel('Accracy')
    plt.xlabel('Epoch')
    plt.legend(['Training data', 'Validation data'], loc=0)
    # plt.show()


def plot_loss(history, title=None):     # Loss Visualization
    # summarize history for loss
    if not isinstance(history, dict):
        history = history.history

    plt.plot(history['loss'])           # loss
    plt.plot(history['val_loss'])       # validation
    if title is not None:
        plt.title(title)
    plt.ylabel('Loss')
    plt.xlabel('Epoch')
    plt.legend(['Training data', 'Validation data'], loc=0)
    # plt.show()
In [0]:
# Visualization
plot_acc(history, '(a) Accuracy')  # 학습 경과에 따른 정확도 변화 추이
plt.show()
plot_loss(history, '(b) Loss')     # 학습 경과에 따른 손실값 변화 추이
plt.show()