[DOC] How to specify class for balanced random forest classifier?

[sandra-tilmon]

Describe the issue linked to the documentation

https://imbalanced-learn.org/stable/references/generated/imblearn.ensemble.BalancedRandomForestClassifier.html

When setting sampling_strategy as 'str', there is no instruction on how to specify the class targeted for resampling. I would like to know how to set the class as 'not majority'.

Suggest a potential alternative/fix

Please provide an example or instructions on how to set the class when sampling_strategy is 'str' for multiple value rebalancing.

Thank you!

[hayesall]

Passing sampling_strategy="not majority" should resample the non majority label(s) in the label vector. These should be determined automatically.

For example, here the non-majority label would be class 0:

from imblearn.ensemble import BalancedRandomForestClassifier
from imblearn.datasets import make_imbalance
from sklearn.datasets import load_breast_cancer

X, y = load_breast_cancer(return_X_y=True)
X_res, y_res = make_imbalance(X, y, sampling_strategy={0: 50})

clf = BalancedRandomForestClassifier(sampling_strategy="not majority")
clf.fit(X_res, y_res)

[glemaitre]

@hayesall is right. Closing.

[sandra-tilmon]

Thanks for the solution, recommend amending the documentation.

[glemaitre]

We already have a full example explaining how to use sampling_strategy: https://imbalanced-learn.org/dev/auto_examples/api/plot_sampling_strategy_usage.html#sphx-glr-auto-examples-api-plot-sampling-strategy-usage-py