Catboost 不能做多变量回归？

问题:

代码

# CatBoost Regression Model
from catboost import CatBoostRegressor

model = CatBoostRegressor(
        iterations = 2000, 
        learning_rate = 0.1,
        l2_leaf_reg = 3,  
        depth = 9,
        rsm = 1,
        border_count=128,
        bagging_temperature= 10,
        verbose=False,
        loss_function='MultiRMSE',
    
                            )
    

model.fit(X_train, y_train)
y_pred = model.predict(X_test)

报错

Currently only multi-regression, multilabel and survival objectives work with multidimensional target 

原因

Catboost 默认用的回归的损失是RMSE，所以针对多变量的回归任务/分类任务需要更改损失函数

解决方法

model = CatBoostRegressor(
        iterations = 2000, 
        learning_rate = 0.1,
        l2_leaf_reg = 3,  
        depth = 9,
        rsm = 1,
        border_count=128,
        bagging_temperature= 10,
        verbose=False,
        loss_function='MultiRMSE', ## 针对多变量任务更改损失函数改成MultiRMSE 或者 MultiLogit
    
                            )