ASAC 50일차_딥러닝 5일차

2024.02.13

 

CNN 실습

* 이전 일차에 이어서 진행

시도 2

overfit 문제 해결 위한 stride, pooling, dropout 적용

model = tf.keras.Sequential([
	tf.keras.layers.Conv2D(input_shape=(28,28,1), kernel_size=(3,3), filters=16),
	tf.keras.layers.MaxPooling2D(pool_size(2,2), strides=(2,2)),
	tf.keras.layers.Conv2D(kernel_size=(3,3), filters = 32),
	tf.keras.layers.MaxPooling2D(pool_size(2,2), strides=(2,2)),
	tf.keras.layers.Conv2D(kernel_size=(3,3), filters = 64),
	tf.kears.layers.Flatten(),
	tf.keras.layers.Dropout(rate = 0.3),
	tf.keras.layers.Dense(units= 128, activation="relu"),
	tf.keras.layers.Dropout(rate = 0.3),
	tf.keras.layers.Dense(units= 10, activation = 'softmax')
])
model.summary()

# epoch 수를 늘리고 callback + earlystopping 적용

import os 

cp_path = 'training/cp--{epoch:04d}.ckpt'
cp_dir = os.path.dirname(cp_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(cp_path, verbose = 1, save_weights_only = True)
es_callback = tf.keras.callbacks.EarlyStopping(monitor = "val_loss", patience = 10)

model = tf.keras.Sequential([
	# 위와 동일
])
model.summary()


# callback 코드 적용
model.compile(
    loss = "sparse_categorical_crossentropy",
    optimizer = tf.keras.optimizers.Adam(),
		metrics = ["accuracy"]
)

history = model.fit(train_X, 
                    train_y, 
                    epochs=20, 
                    validation_split= 0.25, batch_size= 128, 
                    ##############코드 추가 부분 ##############
                    callbacks = [cp_callback, es_callback])


# 이후 최종 결정
load_cp_weights= "training/cp-0025.ckpt"

model.load_weights(load_cp_weights)
model.evaluate(test_X, test_y)