[DL Specialization] C4W1A2

Coursera

[DL Specialization] C4W1A2

andre99 2024. 12. 28. 10:37

The Sequential API

레이어를 직선으로 쌓아서 모델을 구성 (단순한 설계)

Load the Data and Split the Data into Train/Test Sets

Happy House 데이터셋을 활용하여 얼굴 이미지에서 사람의 미소 여부를 판단하는 모델

모델 목표: 웃고 있는 사람만 Happy House에 입장할 수 있도록 웃는 얼굴 분류

X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = load_happy_dataset()

# Normalize image vectors
X_train = X_train_orig/255.
X_test = X_test_orig/255.

# Reshape
Y_train = Y_train_orig.T
Y_test = Y_test_orig.T

number of training examples = 600
number of test examples = 150
X_train shape: (600, 64, 64, 3)
Y_train shape: (600, 1)
X_test shape: (150, 64, 64, 3)
Y_test shape: (150, 1)

happyModel 함수 구현

레이어 구성 및 하이퍼파라미터:

ZeroPadding2D: 패딩 (3, 3) 사용, 입력 형태 (64, 64, 3)
Conv2D: 필터 32개, 크기 (7, 7), 스트라이드 (1, 1)
BatchNormalization: 축(axis) 3
ReLU: 활성화 함수
MaxPooling2D: 기본 매개변수 사용
Flatten: 출력 데이터를 1차원으로 변환
Dense: 유닛 1개, 활성화 함수 sigmoid

from tensorflow.keras.layers import ZeroPadding2D, Conv2D, BatchNormalization, ReLU, MaxPooling2D, Flatten, Dense

def happyModel():
  
    model = tf.keras.Sequential([
            
            ZeroPadding2D(padding=(3, 3), input_shape=(64, 64, 3)),
        
            Conv2D(32, (7, 7), strides=(1, 1)),

            BatchNormalization(axis=3),

            ReLU(),

            MaxPooling2D(pool_size=(2, 2), strides=(2, 2)),

            Flatten(),

            Dense(1, activation='sigmoid')
            
        ])
    
    return model

['ZeroPadding2D', (None, 70, 70, 3), 0, ((3, 3), (3, 3))]
['Conv2D', (None, 64, 64, 32), 4736, 'valid', 'linear', 'GlorotUniform']
['BatchNormalization', (None, 64, 64, 32), 128]
['ReLU', (None, 64, 64, 32), 0]
['MaxPooling2D', (None, 32, 32, 32), 0, (2, 2), (2, 2), 'valid']
['Flatten', (None, 32768), 0]
['Dense', (None, 1), 32769, 'sigmoid']

happy_model.compile(optimizer='adam',
                   loss='binary_crossentropy',
                   metrics=['accuracy'])

happy_model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
zero_padding2d (ZeroPadding2 (None, 70, 70, 3)         0         
_________________________________________________________________
conv2d (Conv2D)              (None, 64, 64, 32)        4736      
_________________________________________________________________
batch_normalization (BatchNo (None, 64, 64, 32)        128       
_________________________________________________________________
re_lu (ReLU)                 (None, 64, 64, 32)        0         
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 32, 32, 32)        0         
_________________________________________________________________
flatten (Flatten)            (None, 32768)             0         
_________________________________________________________________
dense (Dense)                (None, 1)                 32769     
=================================================================
Total params: 37,633
Trainable params: 37,569
Non-trainable params: 64
_________________________________________________________________

Train and Evaluate the Model

happy_model.fit(X_train, Y_train, epochs=10, batch_size=16)

happy_model.evaluate(X_test, Y_test)

[0.1067291796207428, 0.95333331823349]

The Functional API

레이어를 다양한 방식으로 연결하여 복잡한 설계 가능 (그래프 구조)

Load the SIGNS Dataset & Split the Data into Train/Test Sets

X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = load_signs_dataset()

X_train = X_train_orig/255.
X_test = X_test_orig/255.
Y_train = convert_to_one_hot(Y_train_orig, 6).T
Y_test = convert_to_one_hot(Y_test_orig, 6).T

number of training examples = 1080
number of test examples = 120
X_train shape: (1080, 64, 64, 3)
Y_train shape: (1080, 6)
X_test shape: (120, 64, 64, 3)
Y_test shape: (120, 6)

convolutional_model

def convolutional_model(input_shape):

    input_img = tf.keras.Input(shape=input_shape)
    Z1 = Conv2D(8, (4, 4), strides=(1, 1), padding='SAME')(input_img)
   
    A1 = ReLU()(Z1)
    P1 = MaxPooling2D(pool_size=(8, 8), strides=(8, 8), padding='SAME')(A1)
    
    Z2 = Conv2D(16, (2, 2), strides=(1, 1), padding='SAME')(P1)
    A2 = ReLU()(Z2)
    P2 = MaxPooling2D(pool_size=(4, 4), strides=(4, 4), padding='SAME')(A2)
    
    F = Flatten()(P2)
    
    outputs = Dense(6, activation='softmax')(F)
    model = tf.keras.Model(inputs=input_img, outputs=outputs)
    
    return model

conv_model = convolutional_model((64, 64, 3))
conv_model.compile(optimizer='adam',
                  loss='categorical_crossentropy',
                  metrics=['accuracy'])
conv_model.summary()
    
output = [['InputLayer', [(None, 64, 64, 3)], 0],
        ['Conv2D', (None, 64, 64, 8), 392, 'same', 'linear', 'GlorotUniform'],
        ['ReLU', (None, 64, 64, 8), 0],
        ['MaxPooling2D', (None, 8, 8, 8), 0, (8, 8), (8, 8), 'same'],
        ['Conv2D', (None, 8, 8, 16), 528, 'same', 'linear', 'GlorotUniform'],
        ['ReLU', (None, 8, 8, 16), 0],
        ['MaxPooling2D', (None, 2, 2, 16), 0, (4, 4), (4, 4), 'same'],
        ['Flatten', (None, 64), 0],
        ['Dense', (None, 6), 390, 'softmax']]
    
comparator(summary(conv_model), output)

Model: "functional_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 64, 64, 3)]       0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 64, 64, 8)         392       
_________________________________________________________________
re_lu_1 (ReLU)               (None, 64, 64, 8)         0         
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 8, 8, 8)           0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 8, 8, 16)          528       
_________________________________________________________________
re_lu_2 (ReLU)               (None, 8, 8, 16)          0         
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 2, 2, 16)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 64)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 6)                 390       
=================================================================
Total params: 1,310
Trainable params: 1,310
Non-trainable params: 0
_________________________________________________________________

Train the Model

train_dataset = tf.data.Dataset.from_tensor_slices((X_train, Y_train)).batch(64)
test_dataset = tf.data.Dataset.from_tensor_slices((X_test, Y_test)).batch(64)
history = conv_model.fit(train_dataset, epochs=100, validation_data=test_dataset)

History Object

df_loss_acc = pd.DataFrame(history.history)
df_loss= df_loss_acc[['loss','val_loss']]
df_loss.rename(columns={'loss':'train','val_loss':'validation'},inplace=True)
df_acc= df_loss_acc[['accuracy','val_accuracy']]
df_acc.rename(columns={'accuracy':'train','val_accuracy':'validation'},inplace=True)
df_loss.plot(title='Model loss',figsize=(12,8)).set(xlabel='Epoch',ylabel='Loss')
df_acc.plot(title='Model Accuracy',figsize=(12,8)).set(xlabel='Epoch',ylabel='Accuracy')

[Text(0, 0.5, 'Accuracy'), Text(0.5, 0, 'Epoch')]

Model Loss

에포크가 증가함에 따라 학습 데이터의 손실 값은 꾸준히 감소
검증 데이터의 손실 값도 초기에는 감소하다가 에포크가 진행됨에 따라 변화가 줄어듦
검증 손실이 학습 손실과 크게 차이 나지 않는다면 모델이 과적합되지 않았음을 나타냄

Model Accuracy

학습 데이터의 정확도는 에포크가 증가함에 따라 계속 상승
검증 데이터의 정확도는 초기에는 상승하며 학습 효과가 있음 → 일정 에포크 이후에는 더 이상 크게 증가하지 않고 일정한 수준에서 유지됨
검증 정확도가 학습 정확도보다 낮다면 모델이 학습 데이터에 과적합될 가능성을 의미함