第6篇 | 特征工程详解

特征工程是将原始数据转换为更好表达问题本质的过程。特征工程决定了模型的上限而算法只是逼近这个上限。一、特征工程概述特征工程包括特征构造、特征提取、特征选择三个方面。二、特征构造特征构造是根据领域知识创建新特征。代码示例特征构造import numpy as npimport pandas as pdfrom sklearn.preprocessing import PolynomialFeaturesimport matplotlib.pyplot as pltfrom sklearn.decomposition import PCAfrom sklearn.feature_selection import SelectKBest, mutual_info_regressionfrom sklearn.ensemble import RandomForestRegressorfrom sklearn.preprocessing import StandardScalerimport warningswarnings.filterwarnings(ignore)print( * 60)print(第6篇特征工程)print( * 60)# 模拟房价预测数据集np.random.seed(42)n_samples 1000data {area: np.random.uniform(50, 200, n_samples),rooms: np.random.randint(1, 6, n_samples),age: np.random.uniform(0, 30, n_samples),subway: np.random.uniform(0.5, 10, n_samples),schools: np.random.randint(0, 5, n_samples)}df pd.DataFrame(data)# 目标变量df[price] (df[area] * 3 df[rooms] * 5 df[subway] * (-3) df[schools] * 10 -df[age] * 0.5 np.random.randn(n_samples) * 10)print(原始数据:)print(df.head())# 1. 多项式特征print(\n - * 50)print(1. 多项式特征构造)print(- * 50)X df[[area, rooms]].valuespoly PolynomialFeatures(degree2, include_biasFalse)X_poly poly.fit_transform(X)print(f原始特征数: {X.shape[1]})print(f多项式特征数: {X_poly.shape[1]})# 2. 交互特征print(\n - * 50)print(2. 交互特征构造)print(- * 50)df[area_per_room] df[area] / (df[rooms] 1)df[age_subway] df[age] * df[subway]df[score] df[schools] - df[subway] * 0.5print(新增交互特征:)print(df[[area_per_room, age_subway, score]].head())# 3. 领域特征print(\n - * 50)print(3. 领域特定特征)print(- * 50)df[is_new] (df[age] 3).astype(int)df[near_subway] (df[subway] 1).astype(int)df[good_school] (df[schools] 3).astype(int)print(df[[is_new, near_subway, good_school]].head())三、特征提取与降维当特征数量非常多时可以使用PCA进行降维。代码示例PCA降维print(\n * 50)print(主成分分析PCA降维)print( * 50)feature_cols [area, rooms, age, subway, schools,area_per_room, age_subway, score]X df[feature_cols].values# 标准化scaler StandardScaler()X_scaled scaler.fit_transform(X)# PCApca PCA()X_pca pca.fit_transform(X_scaled)# 方差解释比例print(各主成分的方差解释比例:)for i, ratio in enumerate(pca.explained_variance_ratio_[:5]):print(f PC{i1}: {ratio:.4f} ({ratio*100:.2f}%))print(f\n前3个主成分累计解释方差: {sum(pca.explained_variance_ratio_[:3])*100:.2f}%)# 可视化fig, axes plt.subplots(1, 2, figsize(14, 5))axes[0].bar(range(1, len(pca.explained_variance_ratio_) 1),pca.explained_variance_ratio_, alpha0.7)axes[0].plot(range(1, len(pca.explained_variance_ratio_) 1),np.cumsum(pca.explained_variance_ratio_), ro-)axes[0].set_xlabel(Principal Component)axes[0].set_ylabel(Variance Ratio)axes[0].set_title(PCA Variance Explained)axes[0].legend([Cumulative, Individual])scatter axes[1].scatter(X_pca[:, 0], X_pca[:, 1], cdf[price], cmapviridis, alpha0.6)axes[1].set_xlabel(PC1)axes[1].set_ylabel(PC2)axes[1].set_title(PCA Visualization)plt.colorbar(scatter, axaxes[1], labelPrice)plt.tight_layout()plt.savefig(pca_analysis.png, dpi150)plt.show()四、特征选择特征选择是从众多特征中选择最相关的子集。代码示例特征选择print(\n * 50)print(特征选择方法)print( * 50)y df[price].values# 1. 相关性分析print(\n1. 特征与目标变量的相关性:)correlations df.corr()[price].drop(price).sort_values(keyabs, ascendingFalse)print(correlations.round(4))# 2. SelectKBestselector_f SelectKBest(k5)X_f selector_f.fit_transform(X_scaled, y)selected_features [feature_cols[i] for i in selector_f.get_support(indicesTrue)]print(f\n2. SelectKBest选择的特征: {selected_features})# 3. 互信息分析mi_scores mutual_info_regression(X_scaled, y)mi_df pd.DataFrame({feature: feature_cols, mi_score: mi_scores})mi_df mi_df.sort_values(mi_score, ascendingFalse)print(f\n3. 互信息得分:)for _, row in mi_df.iterrows():print(f {row[feature]}: {row[mi_score]:.4f})# 4. 随机森林特征重要性print(\n4. 随机森林特征重要性:)rf RandomForestRegressor(n_estimators100, random_state42)rf.fit(X_scaled, y)for feat, imp in sorted(zip(feature_cols, rf.feature_importances_), keylambda x: x[1], reverseTrue):bar # * int(imp * 50)print(f {feat:15s}: {imp:.4f} {bar})# 可视化fig, ax plt.subplots(figsize(10, 6))ax.barh(mi_df[feature], mi_df[mi_score], colorsteelblue)ax.set_xlabel(Mutual Information Score)ax.set_title(Feature Importance (Mutual Information))plt.tight_layout()plt.savefig(feature_selection.png, dpi150)plt.show()五、总结特征工程是机器学习成功的关键特征构造多项式特征、交互特征、领域知识特征特征提取PCA降维可以减少维度特征选择过滤法、嵌入法各有优缺点好的特征应具备可区分性、简洁性