3

I want to use RandomOverSampler function from imbalanced-learn module to perform oversampling the data with more than two classes. The following is my code with 3 classes:

import numpy as np
from imblearn.over_sampling import RandomOverSampler

data = np.random.randn(30,5)
label = np.random.randint(3, size=30)

ros = RandomOverSampler(random_state=3)
data_res, label_res = ada.fit_sample(data, label)

After running, it returns this warning:

UserWarning: The target type should be binary. warnings.warn('The target type should be binary.')

But the documentation says:

Notes

Supports mutli-class resampling.

Am I missing something to use it for multi class case? If this is only for binary class, is there any other library or module which supports multi class oversampling?

sophros
  • 14,672
  • 11
  • 46
  • 75
starrr
  • 1,013
  • 1
  • 17
  • 48

3 Answers3

2


I faced the same situation yesterday,
and I used conda to install the library,
I found the file -> base.py
it had something different with the newest version on github.

so I git clone the newest version by github
https://github.com/scikit-learn-contrib/imbalanced-learn
and then,
everythings is ok!
u can use multi-class well

林詠翔
  • 21
  • 2
0

You need to update imblearn using :

pip install -U imbalanced-learn

Bani
  • 101
  • 1
  • 1
  • 8
0

RandomOverSampler() works very well for me on a multi-class problem with 9 labels I see on your code that you are using ada oversampler instead of ros that you defined.

BELMALLEM Marouane
  • 1
  • 1