Hyperparameter Tuning (1)

model = create_model(learning_rate=0.001) model.fit(X_train, y_train, epochs=10, batch_size=64, validation_split=0.2)

import numpy as np from sklearn.model_selection import train_test_split from sklearn.utils import shuffle # 가상의 데이터셋 생성 (CIFAR-10과 유사한 데이터셋) num_classes = 10 input_shape = (32, 32, 3) num_samples = 1000 X = np.random.rand(num_samples, *input_shape) # 1000개의 32x32x3 이미지 생성 y = np.random.randint(num_classes, size=num_samples) # 0부터 9까지의 정수 레이블 생성 # 데이터셋 셔플 및 분할 X, y = shuffle(X, y) # 데이터를 섞어줌 X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) # 80% 훈련, 20% 테스트

from tensorflow.keras.applications import ResNet50 from tensorflow.keras.models import Model from tensorflow.keras.layers import Dense, GlobalAveragePooling2D from tensorflow.keras.optimizers import Adam def create_model(learning_rate=0.001): base_model = ResNet50(weights='imagenet', include_top=False, input_shape=input_shape) x = base_model.output x = GlobalAveragePooling2D()(x) # 글로벌 평균 풀링 레이어 추가 x = Dense(1024, activation='relu')(x) # 완전 연결 레이어 추가 predictions = Dense(num_classes, activation='softmax')(x) # 출력 레이어 추가 model = Model(inputs=base_model.input, outputs=predictions) opt = Adam(learning_rate=learning_rate) # Adam 옵티마이저 설정 model.compile(optimizer=opt, loss='sparse_categorical_crossentropy', metrics=['accuracy']) # 모델 컴파일 return model

learning_rates = [0.001, 0.01, 0.1] batch_sizes = [16, 32, 64] epochs = 10 best_accuracy = 0 best_params = {} # 하이퍼파라미터 조합을 수동으로 설정하여 모델 학습 및 평가 for lr in learning_rates: for batch_size in batch_sizes: model = create_model(learning_rate=lr) model.fit(X_train, y_train, epochs=epochs, batch_size=batch_size, validation_split=0.2, verbose=1) loss, accuracy = model.evaluate(X_test, y_test, verbose=0) print(f"Learning Rate: {lr}, Batch Size: {batch_size}, Test Accuracy: {accuracy}") if accuracy > best_accuracy: best_accuracy = accuracy best_params = {'learning_rate': lr, 'batch_size': batch_size} print("Best Parameters:", best_params) print("Best Test Accuracy:", best_accuracy)

# 최적의 하이퍼파라미터로 모델 재학습 epochs = 10 best_model = create_model(learning_rate=best_params['learning_rate']) best_model.fit(X_train, y_train, epochs=epochs, batch_size=best_params['batch_size'], validation_split=0.2) # 테스트 데이터셋으로 모델 평가 test_loss, test_accuracy = best_model.evaluate(X_test, y_test) print(f"Test Accuracy: {test_accuracy:.2f}") print(f"Test Loss: {test_loss:.2f}")

import numpy as np import tensorflow as tf from tensorflow.keras.applications import ResNet50 from tensorflow.keras.models import Model from tensorflow.keras.layers import Dense, GlobalAveragePooling2D from tensorflow.keras.optimizers import Adam from sklearn.model_selection import train_test_split, ParameterSampler from sklearn.utils import shuffle from scipy.stats import uniform

# 가상의 데이터셋 생성 (CIFAR-10과 유사한 데이터셋) num_classes = 10 input_shape = (32, 32, 3) num_samples = 1000 X = np.random.rand(num_samples, *input_shape) y = np.random.randint(num_classes, size=num_samples)

# 데이터셋 셔플 및 분할 X, y = shuffle(X, y) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

def create_model(learning_rate=0.001): base_model = ResNet50(weights='imagenet', include_top=False, input_shape=input_shape) x = base_model.output x = GlobalAveragePooling2D()(x) x = Dense(1024, activation='relu')(x) predictions = Dense(num_classes, activation='softmax')(x) model = Model(inputs=base_model.input, outputs=predictions) opt = Adam(learning_rate=learning_rate) model.compile(optimizer=opt, loss='sparse_categorical_crossentropy', metrics=['accuracy']) return model

# 하이퍼파라미터 분포 정의 param_dist = { 'learning_rate': uniform(0.001, 0.1), 'batch_size': [16, 32, 64] } best_accuracy = 0 best_params = {} n_iter = 5 # 랜덤하게 하이퍼파라미터 조합을 선택하여 시도 for params in ParameterSampler(param_dist, n_iter=n_iter, random_state=42): model = create_model(learning_rate=params['learning_rate']) model.fit(X_train, y_train, epochs=10, batch_size=params['batch_size'], validation_split=0.2, verbose=0) loss, accuracy = model.evaluate(X_test, y_test, verbose=0) print(f"Params: {params}, Test Accuracy: {accuracy}") if accuracy > best_accuracy: best_accuracy = accuracy best_params = params print("Best Parameters (RandomSearch):", best_params) print("Best Test Accuracy (RandomSearch):", best_accuracy)

# 최적의 하이퍼파라미터로 모델 재학습 best_model = create_model(learning_rate=best_params['learning_rate']) best_model.fit(X_train, y_train, epochs=10, batch_size=best_params['batch_size'], validation_split=0.2) # 테스트 데이터셋으로 모델 평가 test_loss, test_accuracy = best_model.evaluate(X_test, y_test) print(f"Test Accuracy: {test_accuracy:.2f}") print(f"Test Loss: {test_loss:.2f}")

from sklearn.model_selection import GridSearchCV, train_test_split from sklearn.ensemble import RandomForestClassifier from sklearn.datasets import make_classification # 데이터 준비 X, y = make_classification(n_samples=1000, n_features=20, n_informative=2, n_redundant=10, random_state=42) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) # 하이퍼파라미터 그리드 설정 param_grid = { 'n_estimators': [50, 100, 200], 'max_depth': [10, 20, 30], 'min_samples_split': [2, 5, 10] } # 랜덤 포레스트 모델 정의 model = RandomForestClassifier(random_state=42) # 그리드 서치 설정 grid_search = GridSearchCV(estimator=model, param_grid=param_grid, cv=5, scoring='accuracy') # 그리드 서치 수행 grid_search.fit(X_train, y_train) # 최적의 하이퍼파라미터 출력 print("Best hyperparameters:", grid_search.best_params_) # 최적의 모델로 테스트 데이터 평가 best_model = grid_search.best_estimator_ train_accuracy_tuned = best_model.score(X_train, y_train) test_accuracy_tuned = best_model.score(X_test, y_test) print(f"Tuned Model - Train Accuracy: {train_accuracy_tuned}, Test Accuracy: {test_accuracy_tuned}")

from sklearn.model_selection import RandomizedSearchCV, train_test_split from sklearn.ensemble import RandomForestClassifier from sklearn.datasets import make_classification from scipy.stats import randint # 데이터 준비 X, y = make_classification(n_samples=1000, n_features=20, n_informative=2, n_redundant=10, random_state=42) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) # 하이퍼파라미터 범위 설정 param_dist = { 'n_estimators': randint(50, 200), 'max_depth': randint(10, 30), 'min_samples_split': randint(2, 10) } # 랜덤 포레스트 모델 정의 model = RandomForestClassifier(random_state=42) # 랜덤 서치 설정 random_search = RandomizedSearchCV(estimator=model, param_distributions=param_dist, n_iter=100, cv=5, scoring='accuracy', random_state=42) # 랜덤 서치 수행 random_search.fit(X_train, y_train) # 최적의 하이퍼파라미터 출력 print("Best hyperparameters:", random_search.best_params_) # 최적의 모델로 테스트 데이터 평가 best_model = random_search.best_estimator_ train_accuracy_tuned = best_model.score(X_train, y_train) test_accuracy_tuned = best_model.score(X_test, y_test) print(f"Tuned Model - Train Accuracy: {train_accuracy_tuned}, Test Accuracy: {test_accuracy_tuned}")