使用imblearn Pipeline报错：'Pipeline'对象无'_check_fit_params'属性

代码示例 #

from imblearn.over_sampling import SMOTE
from imblearn.under_sampling import RandomUnderSampler
from imblearn.pipeline import Pipeline

# 定义特征与目标变量
X = df.drop('infected', axis=1)
y = df['infected']

# 划分训练集与测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# 定义采样策略
over = SMOTE(sampling_strategy=0.5)  # 将少数类过采样至多数类的50%
under = RandomUnderSampler(sampling_strategy=0.8)  # 将多数类欠采样至原规模的80%

pipeline = Pipeline(steps=[('o', over), ('u', under)])

# 执行重采样
X_resampled, y_resampled = pipeline.fit_resample(X_train, y_train)

# 输出新的类别分布
print("Resampled class distribution:", pd.Series(y_resampled).value_counts())

报错信息 #

AttributeError                            Traceback (most recent call last)
Cell In[7], line 19
     16 pipeline = Pipeline(steps=[('o', over), ('u', under)])
     18 # Apply the resampling
---> 19 X_resampled, y_resampled = pipeline.fit_resample(X_train, y_train)
     21 # Show the new class distribution
     22 print("Resampled class distribution:", pd.Series(y_resampled).value_counts())

File ~\anaconda3\Lib\site-packages\imblearn\pipeline.py:372, in Pipeline.fit_resample(self, X, y, **fit_params)
    342 """Fit the model and sample with the final estimator.
    343 
    344 Fits all the transformers/samplers one after the other and
   (...)
    369     Transformed target.
    370 """
    371 self._validate_params()
---> 372 fit_params_steps = self._check_fit_params(**fit_params)
    373 Xt, yt = self._fit(X, y, **fit_params_steps)
    374 last_step = self._final_estimator

AttributeError: 'Pipeline' object has no attribute '_check_fit_params'

解决方案 #

这个错误核心原因是imbalanced-learn与scikit-learn版本不兼容，即使更新所有包也可能出现版本匹配错位的情况，以下是两种解决思路：

1. 强制安装版本匹配的包组合

imbalanced-learn的版本需要和scikit-learn严格对应，比如：

• imbalanced-learn 0.11.x ↔ scikit-learn 1.3.x
• imbalanced-learn 0.10.x ↔ scikit-learn 1.2.x

执行以下命令安装匹配版本（以1.3.x/0.11.x为例）：

pip install -U imbalanced-learn==0.11.0 scikit-learn==1.3.0

或者让pip自动处理依赖匹配：

pip install -U "imbalanced-learn[all]"

2. 绕开imblearn Pipeline，手动依次执行采样

不用Pipeline，直接按顺序调用SMOTE和RandomUnderSampler的fit_resample方法：

# 先执行过采样
X_over, y_over = over.fit_resample(X_train, y_train)
# 再执行欠采样
X_resampled, y_resampled = under.fit_resample(X_over, y_over)

# 输出分布
print("Resampled class distribution:", pd.Series(y_resampled).value_counts())

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