### 特征分析与回归模型构建 #### 数据加载与预处理 在进行特征分析之前，需要先加载数据并对其进行必要的清理和转换。以下是具体操作： ```python import numpy as np import pandas as pd # 加载数据 train = pd.read_csv("路径/train.csv") test = pd.read_csv("路径/test.csv") # 处理缺失值 def handle_missing_data(df): for col in df.columns: if df[col].dtype == 'object': df[col] = df[col].fillna('None') elif df[col].dtype in ['int64', 'float64']: df[col] = df[col].fillna(df[col].median()) return df train = handle_missing_data(train) test = handle_missing_data(test) # 转换目标变量为对数形式以减少偏态分布的影响[^4] y_train = np.log1p(train['SalePrice']) train.drop(['SalePrice'], axis=1, inplace=True) ``` #### 特征工程 通过探索性数据分析 (EDA)，可以发现哪些特征可能对预测有帮助。 ```python import seaborn as sns import matplotlib.pyplot as plt # 可视化数值型特征与目标变量的关系 numeric_features = train.select_dtypes(include=['int64', 'float64']).columns.tolist() fig, axes = plt.subplots(len(numeric_features), 1, figsize=(8, 4 * len(numeric_features))) for i, feature in enumerate(numeric_features): sns.scatterplot(x=train[feature], y=y_train, ax=axes[i]) plt.tight_layout() # 对类别型特征进行编码 categorical_features = train.select_dtypes(include=['object']).columns.tolist() train = pd.get_dummies(train, columns=categorical_features, drop_first=True) test = pd.get_dummies(test, columns=categorical_features, drop_first=True) ``` #### 构建回归模型 为了提高预测性能，可以选择多种机器学习算法，并将其组合成集成方法。 ##### 单一模型训练 以下是一个简单的线性回归模型作为基础示例： ```python from sklearn.linear_model import LinearRegression from sklearn.metrics import mean_squared_error model_lr = LinearRegression() model_lr.fit(train.values, y_train) predictions_lr = model_lr.predict(train.values) print(f'Linear Regression RMSLE: {np.sqrt(mean_squared_error(y_train, predictions_lr))}') ``` ##### 集成模型训练 使用 XGBoost 和 LightGBM 这样的梯度提升树模型通常能获得更优的结果。还可以进一步采用堆叠技术来融合多个模型。 ```python import xgboost as xgb import lightgbm as lgb from mlxtend.regressor import StackingCVRegressor # 定义各个基模型 xgb_reg = xgb.XGBRegressor(objective='reg:squarederror', n_estimators=1000, learning_rate=0.05) lgb_reg = lgb.LGBMRegressor(num_leaves=31, max_depth=-1, random_state=3, importance_type='gain') stacked_regressor = StackingCVRegressor(regressors=[xgb_reg, lgb_reg], meta_regressor=xgb_reg, use_features_in_secondary=True) # 训练最终模型 stacked_regressor.fit(train.values, y_train) predictions_stacked = stacked_regressor.predict(train.values) print(f'Stacked Regressor RMSLE: {np.sqrt(mean_squared_error(y_train, predictions_stacked))}')[^1] ``` #### 测试集预测 最后，在测试集上应用训练好的模型生成预测结果。 ```python final_predictions = np.expm1(stacked_regressor.predict(test.values)) submission = pd.DataFrame({'Id': test.Id, 'SalePrice': final_predictions}) submission.to_csv('submission.csv', index=False) ``` ---