使用cross_val_score设置scoring为neg_mean_squared_error触发TypeError

问题场景 #

在汽车价格预测的回归任务中，使用cross_val_score评估模型时，设置scoring='neg_mean_squared_error'触发如下错误：

TypeError                                 Traceback (most recent call last)
Cell In [46], line 1
----> 1 cross_val_score(model, transformed_X, y, cv=5, scoring='neg_mean_squared_error')

TypeError: 'numpy.float64' object is not callable

注：不设置scoring参数时，np.mean(cross_val_score(model, transformed_X, y, cv=5))可正常运行。

问题代码片段 #

报错核心代码：

cross_val_score(model, transformed_X, y, cv=5, scoring='neg_mean_squared_error')

完整代码：

from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer

OneHotEncoder = OneHotEncoder()
categorical_features = ["Make", "Colour", "Doors"]
transformer = ColumnTransformer([("one-hot",
                                   OneHotEncoder,
                                   categorical_features)],
                                   remainder="passthrough")

transformed_X = transformer.fit_transform(X)
transformed_X

from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split

np.random.seed(0)

X = car_sales.drop("Price", axis=1)
y = car_sales["Price"]

X_train, X_test, y_train, y_test = train_test_split(transformed_X,
                                                    y,
                                                    test_size=0.2)

model = RandomForestRegressor()
model.fit(X_train, y_train)
model.score(X_test, y_test)

cross_val_score(model, transformed_X, y, cv=5, scoring='neg_mean_squared_error')

transformed_X示例：

array([[0.0000000e+00, 1.0000000e+00, 0.0000000e+00, ..., 3.5431000e+04,
        4.3200000e+02, 2.2015860e+04],
       [1.0000000e+00, 0.0000000e+00, 0.0000000e+00, ..., 1.9271400e+05,
        1.0300000e+02, 1.1974723e+05],
       [0.0000000e+00, 1.0000000e+00, 0.0000000e+00, ..., 8.4714000e+04,
        3.3400000e+02, 5.2638970e+04],
       ...,
       [0.0000000e+00, 0.0000000e+00, 1.0000000e+00, ..., 6.6604000e+04,
        4.7300000e+02, 4.1385910e+04],
       [0.0000000e+00, 1.0000000e+00, 0.0000000e+00, ..., 2.1588300e+05,
        1.8000000e+01, 1.3414381e+05],
       [0.0000000e+00, 0.0000000e+00, 0.0000000e+00, ..., 2.4836000e+05,
        5.1000000e+01, 1.5432413e+05]])

问题原因 #

代码存在命名冲突：

• 先导入from sklearn.preprocessing import OneHotEncoder（类对象），随后执行OneHotEncoder = OneHotEncoder()，将原类对象覆盖为类的实例对象。
• 后续ColumnTransformer中使用的OneHotEncoder是实例而非类，导致数据转换流程出现隐藏异常，最终在cross_val_score调用评分指标时触发类型错误。

解决方案 #

修改实例变量名，避免覆盖原类：

from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split, cross_val_score
import numpy as np

# 重命名实例变量，避免覆盖原OneHotEncoder类
one_hot_encoder = OneHotEncoder()
categorical_features = ["Make", "Colour", "Doors"]
transformer = ColumnTransformer([("one-hot",
                                   one_hot_encoder,
                                   categorical_features)],
                                   remainder="passthrough")

np.random.seed(0)

X = car_sales.drop("Price", axis=1)
y = car_sales["Price"]

# 先拆分数据集再做特征转换，避免数据泄露（原代码顺序有误）
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
transformed_X_train = transformer.fit_transform(X_train)
transformed_X_test = transformer.transform(X_test)

model = RandomForestRegressor()
model.fit(transformed_X_train, y_train)
model.score(transformed_X_test, y_test)

# 现在可正常执行
cross_val_score(model, transformer.fit_transform(X), y, cv=5, scoring='neg_mean_squared_error')

额外优化：

• 调整数据处理顺序，先拆分数据集再做特征转换，避免数据泄露。
• 显式导入所有需要的库，保证代码独立性。

内容的提问来源于stack exchange，提问作者Yusuf Saad