성능 향상 using callback

1. 각종 모듈 호출

import numpy as np
from sklearn.model_selection import train_test_split
from tensorflow.keras.datasets import cifar10
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, Activation, MaxPooling2D,\
                                                                         Flatten, Dropout, Dense
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import ModelCheckpoint, ReduceLROnPlateau, EarlyStopping

---

2. 각종 함수 정의

def make_zero_to_one(images, labels) :
    
    images = np.array(images/255., dtype = np.float32)
    labels = np.array(labels, dtype = np.float32)
    
    return images, labels


def ohe(labels) :
    
    labels = to_categorical(labels)
    
    return labels


def tr_val_test(train_images, train_labels, test_images, test_labels, val_rate) :
    
    tr_images, val_images, tr_labels, val_labels = \
                          train_test_split(train_images, train_labels, test_size = val_rate)
    
    return (tr_images, tr_labels), (val_images, val_labels), (test_images, test_labels)


def create_model(images, verbose) :
    
    input_size = (images.shape)[1]
    input_tensor = Input(shape = (input_size, input_size, 3))
    
    x = Conv2D(filters=32, kernel_size=(3, 3), padding='same')(input_tensor)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)

    x = Conv2D(filters=32, kernel_size=(3, 3), padding='same')(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D(pool_size=(2, 2))(x)

    x = Conv2D(filters=64, kernel_size=3, padding='same')(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)

    x = Conv2D(filters=64, kernel_size=3, padding='same')(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D(pool_size=2)(x)

    x = Conv2D(filters=128, kernel_size=3, padding='same')(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)

    x = Conv2D(filters=128, kernel_size=3, padding='same')(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D(pool_size=2)(x)

    # cifar10의 클래스가 10개 이므로 마지막 classification의 Dense layer units갯수는 10
    x = Flatten(name='flatten')(x)
    x = Dropout(rate=0.5)(x)
    x = Dense(300, activation='relu', name='fc1')(x)
    x = Dropout(rate=0.3)(x)
    output = Dense(10, activation='softmax', name='output')(x)

    model = Model(inputs=input_tensor, outputs=output)
    
    if verbose:
        model.summary()
    
    return model

---

3. 데이터 호출 및 전처리

(train_images, train_labels), (test_images, test_labels) = cifar10.load_data()

train_images, train_labels = make_zero_to_one(train_images, train_labels)
test_images, test_labels   = make_zero_to_one(test_images, test_labels)

train_labels = ohe(train_labels)
test_labels  = ohe(test_labels)

(tr_images, tr_labels), (val_images, val_labels), (test_images, test_labels) = \
             tr_val_test(train_images, train_labels, test_images, test_labels, val_rate = 0.15)

---

4. 기존 저장된 모델 삭제

!rm *.hdf5

---

5. 모델 생성 및 학습

model = create_model(tr_images, verbose=True)

model.compile(optimizer = Adam(lr = 0.001), loss = "categorical_crossentropy", metrics = ["accuracy"])

mcp = ModelCheckpoint(filepath = "/kaggle/working/weights.{epoch:02d}-{val_loss:.2f}.hdf5",
                      monitor  = "val_loss",
                      save_best_only    = True,
                      save_weights_only = True,
                      mode    = "min",
                      period  = 1,
                      verbose = False)

rlr = ReduceLROnPlateau(monitors = "val_loss", factor  = 0.2, patience = 5, mode = "min", verbose = True)
ely = EarlyStopping(monitor = "val_loss", patience = 10, mode = "min", verbose = True)

result = model.fit(tr_images, tr_labels, batch_size = 32, epochs = 30, shuffle = True, \
                         validation_data = (val_images, val_labels), callbacks = [mcp, rlr, ely])

---

6. 마지막까지 학습된 모델로 성능 평가

model.evaluate(test_images, test_labels, batch_size = 32)

---

7. 저장된 weight 확인해보기

!ls -lia

---

8. 최적 weight를 모델로 재로딩한 후, 테스트 데이터로 다시 평가하기

model = create_model(tr_images, verbose = True)
model.compile(optimizer = Adam(learning_rate = 0.001), loss = "categorical_crossentropy", \
                                                                     metrics = ["accuracy"])
model.load_weights("/kaggle/working/weights.19-0.53.hdf5")

model.evaluate(test_images, test_labels, batch_size = 32)