import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, models
import numpy as np
import matplotlib.pyplot as plt
from keras.callbacks import EarlyStopping
print(tf.__version__)
print(np.__version__)
2.11.0 1.21.5
fashion_mnist = keras.datasets.fashion_mnist
# 4개의 데이터 셋 반환(numpy 배열)
(X_train, y_train), (X_test, y_test) = fashion_mnist.load_data()
print("학습용 데이터 : x: {}, y:{}".format(X_train.shape, y_train.shape) )
print("테스트 데이터 : x: {}, y:{}".format(X_test.shape, y_test.shape) )
학습용 데이터 : x: (60000, 28, 28), y:(60000,) 테스트 데이터 : x: (10000, 28, 28), y:(10000,)
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
print("학습용 데이터의 레이블 ", np.unique(y_train) )
학습용 데이터의 레이블 [0 1 2 3 4 5 6 7 8 9]
from tensorflow.keras.layers import Conv2D, MaxPool2D, Flatten, Dense
model = models.Sequential()
model.add(Conv2D(32, (3, 3), padding='same',
activation='relu', input_shape=(28, 28, 1)))
model.add(MaxPool2D((2,2)) )
model.add(Conv2D(32, (3,3), padding='same',
strides=1, activation='relu'))
model.add(MaxPool2D(pool_size=2))
# FCL(fully connected layer)
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dense(10, activation='softmax'))
model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_4 (Conv2D) (None, 28, 28, 32) 320
max_pooling2d_4 (MaxPooling (None, 14, 14, 32) 0
2D)
conv2d_5 (Conv2D) (None, 14, 14, 32) 9248
max_pooling2d_5 (MaxPooling (None, 7, 7, 32) 0
2D)
flatten_2 (Flatten) (None, 1568) 0
dense_4 (Dense) (None, 256) 401664
dense_5 (Dense) (None, 10) 2570
=================================================================
Total params: 413,802
Trainable params: 413,802
Non-trainable params: 0
_________________________________________________________________
from keras.callbacks import EarlyStopping, ModelCheckpoint
os.getcwd()
'D:\\GitHub\\DeepLearning_Basic_Class'
early_stopping = EarlyStopping(patience = 30,
monitor="val_loss",
mode="min") # 조기종료 콜백함수 정의
MODEL_SAVE_FOLDER_PATH = "./model/"
if not os.path.exists(MODEL_SAVE_FOLDER_PATH):
os.mkdir(MODEL_SAVE_FOLDER_PATH)
model_path = MODEL_SAVE_FOLDER_PATH + "{epoch:02d}_{val_loss:.4f}.hdf5"
model_path
'./model/{epoch:02d}-{val_loss:.4f}.hdf5'
# checkpoint_path : 모델을 저장할 경로
# monitor : 모델을 저장할 때, 기준이 되는 값
# verbose : 0 (모델이 저장될 경우, 저장), 1 (화면에 표시없이 저장)
# save_best_only : True(모니터 되는 값을 기준으로 가장 좋은 값으로 저장)
# , False(매 에폭마타 모델이 저장.)
# mode : 'auto', 'min', 'max'
checkpoint = ModelCheckpoint(filepath = model_path, monitor='val_loss',
verbose=1,
save_best_only=True, mode='min')
callbacks_list = [checkpoint, early_stopping]
%%time
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
hist = model.fit(X_train, y_train, validation_data=(X_test, y_test),
epochs=3, batch_size=32,
callbacks=callbacks_list )
Epoch 1/3 1874/1875 [============================>.] - ETA: 0s - loss: 0.2374 - accuracy: 0.9125 Epoch 1: val_loss improved from inf to 0.31792, saving model to ./model\01-0.3179.hdf5 1875/1875 [==============================] - 64s 34ms/step - loss: 0.2374 - accuracy: 0.9125 - val_loss: 0.3179 - val_accuracy: 0.8844 Epoch 2/3 1875/1875 [==============================] - ETA: 0s - loss: 0.2028 - accuracy: 0.9252 Epoch 2: val_loss did not improve from 0.31792 1875/1875 [==============================] - 65s 34ms/step - loss: 0.2028 - accuracy: 0.9252 - val_loss: 0.3345 - val_accuracy: 0.8883 Epoch 3/3 1875/1875 [==============================] - ETA: 0s - loss: 0.1840 - accuracy: 0.9316 Epoch 3: val_loss improved from 0.31792 to 0.29546, saving model to ./model\03-0.2955.hdf5 1875/1875 [==============================] - 70s 37ms/step - loss: 0.1840 - accuracy: 0.9316 - val_loss: 0.2955 - val_accuracy: 0.9000 Wall time: 3min 18s
def create_model():
model = models.Sequential()
model.add(Conv2D(32, (3, 3), padding='same',
activation='relu', input_shape=(28, 28, 1)))
model.add(MaxPool2D((2,2)) )
model.add(Conv2D(32, (3,3), padding='same',
strides=1, activation='relu'))
model.add(MaxPool2D(pool_size=2))
# FCL(fully connected layer)
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dense(10, activation='softmax'))
return model
# 기본 모델 instance를 생성
model_hdf5 = create_model()
model_path_hdf5 = os.getcwd() + "/model/" + "03_0.2955.hdf5"
model_path_hdf5
'D:\\GitHub\\DeepLearning_Basic_Class/model/03_0.2955.hdf5'
model_hdf5.load_weights(model_path_hdf5)
model_hdf5.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
score = model_hdf5.evaluate(X_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
Test loss: 0.29545581340789795 Test accuracy: 0.8999999761581421