使用CatBoost进行多目标回归时出现维度错误问题求助

问题场景 #

使用CatBoost进行多目标回归时，执行model.fit()触发如下错误：

Attempt to use multi-dimensional target as one-dimensional

原实现代码：

target_col = ['SISU_LP', 'SISU_AM', 'SISU_EP', 'SISU_HR', 'SISU_HS', 'SISU_HO']
text_cols=['surrender','feelings']
X = data.drop(columns=target_col)
y = data[target_col]
params = {'learning_rate': 0.1, 'depth': 6, 
      'loss_function': 'MultiRMSE',  'eval_metric': 'MultiRMSE'}
model = CatBoostRegressor(**params)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=1)
pool_train = Pool(data=X_train, label=y_train,text_features = text_cols)
pool_test = Pool(data=X_test, label=y_test, text_features = text_cols)
model.fit(pool_train, eval_set=pool_test, use_best_model=True)

报错原因 #

CatBoost的Pool对象在处理多目标变量时，若传入的label是pandas DataFrame格式，部分版本会无法正确识别其多维结构，误判为一维目标。

解决方法 #

1. 将目标变量转为numpy数组 #

创建Pool时，把y_train和y_test转换为numpy数组格式，替换原label参数：

pool_train = Pool(data=X_train, label=y_train.to_numpy(), text_features=text_cols)
pool_test = Pool(data=X_test, label=y_test.to_numpy(), text_features=text_cols)

2. 升级CatBoost版本 #

若上述方法无效，可能是旧版本对多目标DataFrame支持不足，执行升级命令：

pip install --upgrade catboost

修改后完整代码 #

target_col = ['SISU_LP', 'SISU_AM', 'SISU_EP', 'SISU_HR', 'SISU_HS', 'SISU_HO']
text_cols=['surrender','feelings']
X = data.drop(columns=target_col)
y = data[target_col]
params = {'learning_rate': 0.1, 'depth': 6, 
      'loss_function': 'MultiRMSE',  'eval_metric': 'MultiRMSE'}
model = CatBoostRegressor(**params)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=1)
# 转换目标为numpy数组
pool_train = Pool(data=X_train, label=y_train.to_numpy(), text_features = text_cols)
pool_test = Pool(data=X_test, label=y_test.to_numpy(), text_features = text_cols)
model.fit(pool_train, eval_set=pool_test, use_best_model=True)

内容的提问来源于stack exchange，提问作者user3719749