现在有catboost和lgbm模型，如何集成两者 提高预测准确率，请给出python代码

可以使用两种集成方法：Stacking和Blending。

Stacking：将catboost和lgbm的预测结果作为特征输入到另一个机器学习模型中，例如XGBoost、Random Forest等。

下面是一个基本的Stacking代码示例：

from sklearn.model_selection import KFold
from sklearn.metrics import mean_squared_error
from catboost import CatBoostRegressor
import lightgbm as lgb
import numpy as np

# Load data and split into training and validation sets
X_train, X_val, y_train, y_val = ...

# Create base models
catboost = CatBoostRegressor(...)
lgbm = lgb.LGBMRegressor(...)

# Train base models on training set
catboost.fit(X_train, y_train)
lgbm.fit(X_train, y_train)

# Generate predictions on validation set
catboost_preds = catboost.predict(X_val)
lgbm_preds = lgbm.predict(X_val)

# Create new features with base model predictions
X_val_new = np.column_stack((catboost_preds, lgbm_preds))

# Train meta model on new features and true labels
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
meta_model = XGBRegressor(...)
for train_idx, val_idx in kfold.split(X_val_new):
    X_train_fold, y_train_fold = X_val_new[train_idx], y_val[train_idx]
    X_val_fold, y_val_fold = X_val_new[val_idx], y_val[val_idx]
    meta_model.fit(X_train_fold, y_train_fold)

# Generate final predictions on test set
catboost_test_preds = catboost.predict(X_test)
lgbm_test_preds = lgbm.predict(X_test)
X_test_new = np.column_stack((catboost_test_preds, lgbm_test_preds))
final_preds = meta_model.predict(X_test_new)

Blending：将catboost和lgbm的预测结果加权平均，作为最终预测结果。

下面是一个基本的Blending代码示例：

from sklearn.metrics import mean_squared_error
from catboost import CatBoostRegressor
import lightgbm as lgb
import numpy as np

# Load data and split into training and validation sets
X_train, X_val, y_train, y_val = ...

# Create base models
catboost = CatBoostRegressor(...)
lgbm = lgb.LGBMRegressor(...)

# Train base models on training set
catboost.fit(X_train, y_train)
lgbm.fit(X_train, y_train)

# Generate predictions on validation set
catboost_preds = catboost.predict(X_val)
lgbm_preds = lgbm.predict(X_val)

# Combine predictions with weights
final_preds = 0.5 * catboost_preds + 0.5 * lgbm_preds

# Calculate RMSE
rmse = np.sqrt(mean_squared_error(y_val, final_preds))