머신 러닝 (7) Regularization

 import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

import koreanize_matplotlib

import seaborn as sns

from sklearn.datasets import load_diabetes

from sklearn.model_selection import train_test_split

from sklearn.linear_model import LinearRegression

from sklearn.preprocessing import StandardScaler

from sklearn.metrics import mean_squared_error

from sklearn.linear_model import Ridge, Lasso, ElasticNet

X, y = load_diabetes(return_X_y=True)

# 학습용/테스트용 데이터 분할 (8:2)

X_train, X_test, y_train, y_test = train_test_split(

    X, y, test_size=0.2, random_state=42

)

scaler = StandardScaler()

# 학습 데이터 기준으로 스케일러 학습 및 변환

X_train_scaled = scaler.fit_transform(X_train)

# 테스트 데이터는 학습 데이터의 기준(mean, std)으로 변환만 수행

X_test_scaled = scaler.transform(X_test)

ridge = Ridge(alpha=1.0)

ridge.fit(X_train_scaled, y_train)

ridge_pred = ridge.predict(X_test_scaled)

ridge_mse = mean_squared_error(y_test, ridge_pred)

print(f"Ridge MSE: {ridge_mse:.2f}")

print(f"Ridge 회귀 계수:\n{np.round(ridge.coef_, 2)}")

print("-> 모든 변수의 계수가 0이 되지 않고 유지됨 (Shrinkage)")

lasso = Lasso(alpha=1.0)

lasso.fit(X_train_scaled, y_train)

lasso_pred = lasso.predict(X_test_scaled)

lasso_mse = mean_squared_error(y_test, lasso_pred)

lasso_selected_features = np.sum(lasso.coef_ != 0)

print(f"Lasso MSE: {lasso_mse:.2f}")

print(f"Lasso 회귀 계수:\n{np.round(lasso.coef_, 2)}")

print(f"Lasso가 선택한 변수 개수: {lasso_selected_features} / {X.shape[1]}")

print("-> 일부 변수의 계수가 정확히 0이 됨 (Feature Selection)")

# alpha: 규제 강도, l1_ratio: L1 규제의 비율 (0~1 사이)

elastic_net = ElasticNet(alpha=0.1, l1_ratio=0.5, random_state=42)

elastic_net.fit(X_train_scaled, y_train)

el_pred = elastic_net.predict(X_test_scaled)

el_mse = mean_squared_error(y_test, el_pred)

print(f"ElasticNet(Basic) MSE: {el_mse:.2f}")

print(f"ElasticNet 회귀 계수:\n{np.round(elastic_net.coef_, 2)}")

feature_names = load_diabetes().feature_names

plt.figure(figsize=(6, 4))

plt.plot(feature_names, ridge.coef_, 's-', label='Ridge (L2)')

plt.plot(feature_names, lasso.coef_, 'x-', label='Lasso (L1)')

plt.plot(feature_names, elastic_net.coef_, 'o-', label='ElasticNet')

plt.axhline(0, color='black', linestyle='--', linewidth=0.8)

plt.title("Regularization 회귀 계수 비교")

plt.xlabel("Features")

plt.ylabel("Coefficient Value")

plt.legend()

plt.show()