使用imblearn在击打后执行随机欠采样

Question

我正在尝试用RandomUnderSampler()和SMOTE()来实现过采样和欠采样的结合.

我正在处理loan_status数据集。

我已经做了以下的分裂。

X = df.drop(['Loan_Status'],axis=1).values   # independant features
y = df['Loan_Status'].values# dependant variable

这就是我训练数据的分布情况。

​

​

这是我试图执行的用于类平衡的代码片段。

from imblearn.over_sampling import SMOTE
from imblearn.under_sampling import RandomUnderSampler
from imblearn.pipeline import make_pipeline
over = SMOTE(sampling_strategy=0.1)
under = RandomUnderSampler(sampling_strategy=0.5)
pipeline = make_pipeline(over,under)
    
x_sm,y_sm = pipeline.fit_resample(X_train,y_train)

它给了我一个具有以下追溯功能的ValueError：

ValueError                                Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_64588/3438707951.py in <module>
      4 pipeline = make_pipeline(over,under)
      5 
----> 6 x_copy,y_copy = pipeline.fit_resample(x_train_copy,y_train_copy)

~\Anaconda3\lib\site-packages\imblearn\pipeline.py in fit_resample(self, X, y, **fit_params)
    351             fit_params_last_step = fit_params_steps[self.steps[-1][0]]
    352             if hasattr(last_step, "fit_resample"):
--> 353                 return last_step.fit_resample(Xt, yt, **fit_params_last_step)
    354 
    355     @if_delegate_has_method(delegate="_final_estimator")

~\Anaconda3\lib\site-packages\imblearn\base.py in fit_resample(self, X, y)
     77         X, y, binarize_y = self._check_X_y(X, y)
     78 
---> 79         self.sampling_strategy_ = check_sampling_strategy(
     80             self.sampling_strategy, y, self._sampling_type
     81         )

~\Anaconda3\lib\site-packages\imblearn\utils\_validation.py in check_sampling_strategy(sampling_strategy, y, sampling_type, **kwargs)
    532         return OrderedDict(
    533             sorted(
--> 534                 _sampling_strategy_float(sampling_strategy, y, sampling_type).items()
    535             )
    536         )

~\Anaconda3\lib\site-packages\imblearn\utils\_validation.py in _sampling_strategy_float(sampling_strategy, y, sampling_type)
    391             ]
    392         ):
--> 393             raise ValueError(
    394                 "The specified ratio required to generate new "
    395                 "sample in the majority class while trying to "

ValueError: The specified ratio required to generate new sample in the majority class while trying to remove samples. Please increase the ratio.

Answer 1

您必须增加SMOTE的抽样策略，因为((y_train==0).sum())/((y_train==1).sum())高于0.1。看来你的起始不平衡比率是(从眼睛上看) 0.4。尝试：

over = SMOTE(sampling_strategy=0.5)

最后，您可能需要一个相同的最终比率(在低采样之后)，因此您应该将采样策略设置为1.0，用于RandomUnderSampler。

under = RandomUnderSampler(sampling_strategy=1)

试着这样做，如果你有其他问题，给我一个反馈。