-
TensorFlow 기초 25 - MNIST로 CNN 처리TensorFlow 2022. 12. 7. 12:03
# MNIST로 CNN 처리 # 1) Conv(이미지 특징 추출) + Pooling(Conv 결과를 샘플링 - Conv 결과인 Feature map의 크기를 다시 줄임) # 2) 원래의 이미지 크기를 줄여 최종적으로 FCLayer를 진행(Conv + Pooling 결과 다차원 배열 자료를 1차원으로 만들어 한 줄로 세우기) # 3) Dense 층으로 넘겨 분류 작업 수행 import tensorflow as tf from keras import datasets, layers, models (x_train, y_train), (x_test, y_test) = datasets.mnist.load_data() print(x_train.shape, y_train.shape, x_test.shape, y_test.shape) # (60000, 28, 28) (60000,) (10000, 28, 28) (10000,) # CNN은 채널을 사용하기 때문에 3차원 데이터를 4차원으로 변경 x_train = x_train.reshape((60000, 28, 28, 1)) # 흑백은 channel이 1개 x_test = x_test.reshape((10000, 28, 28, 1)) # 예) x_test[3, 12, 13, 1] # print(x_train.ndim) # print(x_train[:1]) x_train = x_train / 255.0 x_test = x_test / 255.0 # print(x_train[:1]) # label의 원핫 처리는 model에게 위임 # model input_shape = (28, 28, 1) # Sequential api 사용 model = models.Sequential() model.add(layers.Conv2D(filters=16, kernel_size=(3, 3), strides=(1, 1), padding='valid', activation='relu', input_shape=input_shape)) model.add(layers.MaxPool2D(pool_size=(2, 2))) model.add(layers.Dropout(rate=0.2)) model.add(layers.Conv2D(filters=32, kernel_size=(3, 3), strides=(1, 1), padding='valid', activation='relu')) model.add(layers.MaxPool2D(pool_size=(2, 2))) # filters는 Dense의 unit과 비슷한 개념이다. model.add(layers.Dropout(rate=0.2)) model.add(layers.Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1), padding='valid', activation='relu')) model.add(layers.MaxPool2D(pool_size=(2, 2))) # kernel_size 명은 생략 가능 model.add(layers.Dropout(rate=0.2)) model.add(layers.Flatten()) # FCLayer(Fully Connected Layer) : 2차원 1차원으로 변경(모든 배열자료를 한 줄로 세움) model.add(layers.Dense(64, activation='relu')) model.add(layers.Dropout(rate=0.2)) model.add(layers.Dense(32, activation='relu')) model.add(layers.Dropout(rate=0.2)) model.add(layers.Dense(10, activation='softmax')) print(model.summary()) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\ from keras.callbacks import EarlyStopping es = EarlyStopping(monitor='val_loss', patience=3) # patience는 많이 줘야됨 history = model.fit(x_train, y_train, batch_size=128, epochs=1000, verbose=0, validation_split=0.2, callbacks=[es]) # history 저장 import pickle history = history.history with open('cnn2_history.pickle', 'wb') as f: pickle.dump(history, f) # 모델 평가 train_loss, train_acc = model.evaluate(x_train, y_train) test_loss, test_acc = model.evaluate(x_test, y_test) print('train_loss : {}, train_acc : {}'.format(train_loss, train_acc)) print('test_loss : {}, test_acc : {}'.format(test_loss, test_acc)) # 모델 저장 model.save('cnn2_model.h5') print() # --- 학습된 모델로 작업 --- mymodel = tf.keras.models.load_model('cnn2_model.h5') # predict import numpy as np print('예측값 :', np.argmax(mymodel.predict(x_test[:1]))) print('예측값 :', np.argmax(mymodel.predict(x_test[[0]]))) # 위랑 같은 의미 print('실제값 :', y_test[0]) # 시각화 import matplotlib.pyplot as plt with open('cnn2_history.pickle', 'rb') as f: history = pickle.load(f) def plot_acc_func(title=None): plt.plot(history['accuracy'], label='accuracy') plt.plot(history['val_accuracy'], label='val_accuracy') plt.title(title) plt.xlabel('epochs') plt.ylabel(title) plt.legend() plot_acc_func('accuracy') plt.show() def plot_loss_func(title=None): plt.plot(history['loss'], label='loss') plt.plot(history['val_loss'], label='val_loss') plt.title(title) plt.xlabel('epochs') plt.ylabel(title) plt.legend() plot_acc_func('loss') plt.show() <console> (60000, 28, 28) (60000,) (10000, 28, 28) (10000,) Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d (Conv2D) (None, 26, 26, 16) 160 max_pooling2d (MaxPooling2D (None, 13, 13, 16) 0 ) dropout (Dropout) (None, 13, 13, 16) 0 conv2d_1 (Conv2D) (None, 11, 11, 32) 4640 max_pooling2d_1 (MaxPooling (None, 5, 5, 32) 0 2D) dropout_1 (Dropout) (None, 5, 5, 32) 0 conv2d_2 (Conv2D) (None, 3, 3, 64) 18496 max_pooling2d_2 (MaxPooling (None, 1, 1, 64) 0 2D) dropout_2 (Dropout) (None, 1, 1, 64) 0 flatten (Flatten) (None, 64) 0 dense (Dense) (None, 64) 4160 dropout_3 (Dropout) (None, 64) 0 dense_1 (Dense) (None, 32) 2080 dropout_4 (Dropout) (None, 32) 0 dense_2 (Dense) (None, 10) 330 ================================================================= Total params: 29,866 Trainable params: 29,866 Non-trainable params: 0 _________________________________________________________________ None 1/1875 [..............................] - ETA: 31s - loss: 0.0028 - accuracy: 1.0000 21/1875 [..............................] - ETA: 4s - loss: 0.0261 - accuracy: 0.9896 40/1875 [..............................] - ETA: 4s - loss: 0.0370 - accuracy: 0.9883 58/1875 [..............................] - ETA: 4s - loss: 0.0316 - accuracy: 0.9914 77/1875 [>.............................] - ETA: 4s - loss: 0.0297 - accuracy: 0.9919 96/1875 [>.............................] - ETA: 4s - loss: 0.0343 - accuracy: 0.9906 114/1875 [>.............................] - ETA: 4s - loss: 0.0309 - accuracy: 0.9915 134/1875 [=>............................] - ETA: 4s - loss: 0.0294 - accuracy: 0.9916 154/1875 [=>............................] - ETA: 4s - loss: 0.0280 - accuracy: 0.9919 174/1875 [=>............................] - ETA: 4s - loss: 0.0280 - accuracy: 0.9916 193/1875 [==>...........................] - ETA: 4s - loss: 0.0275 - accuracy: 0.9916 211/1875 [==>...........................] - ETA: 4s - loss: 0.0273 - accuracy: 0.9917 227/1875 [==>...........................] - ETA: 4s - loss: 0.0277 - accuracy: 0.9916 246/1875 [==>...........................] - ETA: 4s - loss: 0.0270 - accuracy: 0.9916 266/1875 [===>..........................] - ETA: 4s - loss: 0.0281 - accuracy: 0.9914 285/1875 [===>..........................] - ETA: 4s - loss: 0.0290 - accuracy: 0.9911 305/1875 [===>..........................] - ETA: 4s - loss: 0.0292 - accuracy: 0.9912 325/1875 [====>.........................] - ETA: 4s - loss: 0.0287 - accuracy: 0.9914 343/1875 [====>.........................] - ETA: 4s - loss: 0.0288 - accuracy: 0.9916 362/1875 [====>.........................] - ETA: 4s - loss: 0.0284 - accuracy: 0.9920 381/1875 [=====>........................] - ETA: 4s - loss: 0.0283 - accuracy: 0.9920 400/1875 [=====>........................] - ETA: 4s - loss: 0.0277 - accuracy: 0.9921 419/1875 [=====>........................] - ETA: 3s - loss: 0.0276 - accuracy: 0.9920 438/1875 [======>.......................] - ETA: 3s - loss: 0.0278 - accuracy: 0.9919 457/1875 [======>.......................] - ETA: 3s - loss: 0.0275 - accuracy: 0.9920 477/1875 [======>.......................] - ETA: 3s - loss: 0.0268 - accuracy: 0.9921 497/1875 [======>.......................] - ETA: 3s - loss: 0.0271 - accuracy: 0.9922 517/1875 [=======>......................] - ETA: 3s - loss: 0.0267 - accuracy: 0.9923 538/1875 [=======>......................] - ETA: 3s - loss: 0.0266 - accuracy: 0.9923 558/1875 [=======>......................] - ETA: 3s - loss: 0.0267 - accuracy: 0.9924 577/1875 [========>.....................] - ETA: 3s - loss: 0.0265 - accuracy: 0.9925 597/1875 [========>.....................] - ETA: 3s - loss: 0.0258 - accuracy: 0.9928 617/1875 [========>.....................] - ETA: 3s - loss: 0.0256 - accuracy: 0.9927 637/1875 [=========>....................] - ETA: 3s - loss: 0.0262 - accuracy: 0.9925 658/1875 [=========>....................] - ETA: 3s - loss: 0.0261 - accuracy: 0.9925 678/1875 [=========>....................] - ETA: 3s - loss: 0.0260 - accuracy: 0.9923 699/1875 [==========>...................] - ETA: 3s - loss: 0.0255 - accuracy: 0.9925 719/1875 [==========>...................] - ETA: 3s - loss: 0.0256 - accuracy: 0.9925 738/1875 [==========>...................] - ETA: 3s - loss: 0.0253 - accuracy: 0.9925 759/1875 [===========>..................] - ETA: 2s - loss: 0.0254 - accuracy: 0.9925 779/1875 [===========>..................] - ETA: 2s - loss: 0.0258 - accuracy: 0.9925 800/1875 [===========>..................] - ETA: 2s - loss: 0.0256 - accuracy: 0.9926 820/1875 [============>.................] - ETA: 2s - loss: 0.0254 - accuracy: 0.9926 840/1875 [============>.................] - ETA: 2s - loss: 0.0259 - accuracy: 0.9924 861/1875 [============>.................] - ETA: 2s - loss: 0.0260 - accuracy: 0.9924 881/1875 [=============>................] - ETA: 2s - loss: 0.0258 - accuracy: 0.9923 901/1875 [=============>................] - ETA: 2s - loss: 0.0262 - accuracy: 0.9923 922/1875 [=============>................] - ETA: 2s - loss: 0.0259 - accuracy: 0.9923 941/1875 [==============>...............] - ETA: 2s - loss: 0.0257 - accuracy: 0.9925 961/1875 [==============>...............] - ETA: 2s - loss: 0.0254 - accuracy: 0.9925 982/1875 [==============>...............] - ETA: 2s - loss: 0.0252 - accuracy: 0.9926 1003/1875 [===============>..............] - ETA: 2s - loss: 0.0251 - accuracy: 0.9926 1023/1875 [===============>..............] - ETA: 2s - loss: 0.0252 - accuracy: 0.9926 1043/1875 [===============>..............] - ETA: 2s - loss: 0.0250 - accuracy: 0.9927 1062/1875 [===============>..............] - ETA: 2s - loss: 0.0250 - accuracy: 0.9927 1082/1875 [================>.............] - ETA: 2s - loss: 0.0251 - accuracy: 0.9927 1103/1875 [================>.............] - ETA: 2s - loss: 0.0252 - accuracy: 0.9927 1124/1875 [================>.............] - ETA: 1s - loss: 0.0253 - accuracy: 0.9927 1145/1875 [=================>............] - ETA: 1s - loss: 0.0252 - accuracy: 0.9927 1166/1875 [=================>............] - ETA: 1s - loss: 0.0254 - accuracy: 0.9927 1185/1875 [=================>............] - ETA: 1s - loss: 0.0256 - accuracy: 0.9926 1204/1875 [==================>...........] - ETA: 1s - loss: 0.0260 - accuracy: 0.9925 1224/1875 [==================>...........] - ETA: 1s - loss: 0.0257 - accuracy: 0.9925 1245/1875 [==================>...........] - ETA: 1s - loss: 0.0261 - accuracy: 0.9924 1265/1875 [===================>..........] - ETA: 1s - loss: 0.0261 - accuracy: 0.9924 1285/1875 [===================>..........] - ETA: 1s - loss: 0.0262 - accuracy: 0.9924 1305/1875 [===================>..........] - ETA: 1s - loss: 0.0265 - accuracy: 0.9923 1324/1875 [====================>.........] - ETA: 1s - loss: 0.0264 - accuracy: 0.9924 1345/1875 [====================>.........] - ETA: 1s - loss: 0.0263 - accuracy: 0.9924 1366/1875 [====================>.........] - ETA: 1s - loss: 0.0264 - accuracy: 0.9924 1387/1875 [=====================>........] - ETA: 1s - loss: 0.0262 - accuracy: 0.9924 1408/1875 [=====================>........] - ETA: 1s - loss: 0.0260 - accuracy: 0.9925 1428/1875 [=====================>........] - ETA: 1s - loss: 0.0260 - accuracy: 0.9925 1449/1875 [======================>.......] - ETA: 1s - loss: 0.0262 - accuracy: 0.9924 1470/1875 [======================>.......] - ETA: 1s - loss: 0.0263 - accuracy: 0.9924 1491/1875 [======================>.......] - ETA: 0s - loss: 0.0264 - accuracy: 0.9924 1511/1875 [=======================>......] - ETA: 0s - loss: 0.0264 - accuracy: 0.9923 1532/1875 [=======================>......] - ETA: 0s - loss: 0.0265 - accuracy: 0.9923 1552/1875 [=======================>......] - ETA: 0s - loss: 0.0270 - accuracy: 0.9921 1572/1875 [========================>.....] - ETA: 0s - loss: 0.0273 - accuracy: 0.9920 1593/1875 [========================>.....] - ETA: 0s - loss: 0.0277 - accuracy: 0.9918 1613/1875 [========================>.....] - ETA: 0s - loss: 0.0278 - accuracy: 0.9918 1634/1875 [=========================>....] - ETA: 0s - loss: 0.0283 - accuracy: 0.9917 1655/1875 [=========================>....] - ETA: 0s - loss: 0.0287 - accuracy: 0.9915 1675/1875 [=========================>....] - ETA: 0s - loss: 0.0291 - accuracy: 0.9915 1694/1875 [==========================>...] - ETA: 0s - loss: 0.0294 - accuracy: 0.9914 1715/1875 [==========================>...] - ETA: 0s - loss: 0.0295 - accuracy: 0.9914 1735/1875 [==========================>...] - ETA: 0s - loss: 0.0296 - accuracy: 0.9914 1756/1875 [===========================>..] - ETA: 0s - loss: 0.0296 - accuracy: 0.9914 1777/1875 [===========================>..] - ETA: 0s - loss: 0.0299 - accuracy: 0.9913 1798/1875 [===========================>..] - ETA: 0s - loss: 0.0300 - accuracy: 0.9913 1817/1875 [============================>.] - ETA: 0s - loss: 0.0301 - accuracy: 0.9912 1838/1875 [============================>.] - ETA: 0s - loss: 0.0299 - accuracy: 0.9913 1858/1875 [============================>.] - ETA: 0s - loss: 0.0296 - accuracy: 0.9914 1875/1875 [==============================] - 5s 3ms/step - loss: 0.0301 - accuracy: 0.9913 1/313 [..............................] - ETA: 5s - loss: 0.0028 - accuracy: 1.0000 20/313 [>.............................] - ETA: 0s - loss: 0.0270 - accuracy: 0.9891 39/313 [==>...........................] - ETA: 0s - loss: 0.0525 - accuracy: 0.9832 59/313 [====>.........................] - ETA: 0s - loss: 0.0641 - accuracy: 0.9831 79/313 [======>.......................] - ETA: 0s - loss: 0.0722 - accuracy: 0.9818 99/313 [========>.....................] - ETA: 0s - loss: 0.0713 - accuracy: 0.9830 119/313 [==========>...................] - ETA: 0s - loss: 0.0666 - accuracy: 0.9835 138/313 [============>.................] - ETA: 0s - loss: 0.0660 - accuracy: 0.9832 157/313 [==============>...............] - ETA: 0s - loss: 0.0661 - accuracy: 0.9825 177/313 [===============>..............] - ETA: 0s - loss: 0.0598 - accuracy: 0.9839 197/313 [=================>............] - ETA: 0s - loss: 0.0562 - accuracy: 0.9848 217/313 [===================>..........] - ETA: 0s - loss: 0.0582 - accuracy: 0.9847 237/313 [=====================>........] - ETA: 0s - loss: 0.0541 - accuracy: 0.9856 256/313 [=======================>......] - ETA: 0s - loss: 0.0505 - accuracy: 0.9866 275/313 [=========================>....] - ETA: 0s - loss: 0.0483 - accuracy: 0.9869 294/313 [===========================>..] - ETA: 0s - loss: 0.0457 - accuracy: 0.9877 313/313 [==============================] - 1s 3ms/step - loss: 0.0486 - accuracy: 0.9870 train_loss : 0.030139632523059845, train_acc : 0.9913166761398315 test_loss : 0.04862738028168678, test_acc : 0.9869999885559082 1/1 [==============================] - ETA: 0s 1/1 [==============================] - 0s 92ms/step 예측값 : 7 1/1 [==============================] - ETA: 0s 1/1 [==============================] - 0s 13ms/step 예측값 : 7 실제값 : 7CNN은 채널을 사용하기 때문에 3차원 그 전에 데이터를 4차원으로 변경해야 된다.(예) x_test[3, 12, 13, 1])
label의 원핫 처리는 model에게 위임하였다.
accuracy 시각화 
loss 시각화 'TensorFlow' 카테고리의 다른 글