Create a meta-estimator using BalanceCascade sampler

[glemaitre]

As the BalancedBaggingEnsemble, it could be possible to create a meta-estimator using the BalanceEnsemble sampler.

[pythonmite]

I can work on this issue

[chkoar]

@pythonmite go ahead

[pythonmite]

@glemaitre I was trying to figure out the issue in order to implement. Do we need an meta-estimator which would take undersampling or oversampling methods to provide an balanced class sets.

[solegalli]

Hello, by any chance is the algorithm in this article the BalanceCascade in question?

1. make a bag with random undersampling from original dataset
2. train adaboost in the bag
3. remove from original dataset all observations from majority class correctly classified by the adaboost
4. repeat 1-3 with the new "original dataset"

I could try and give it a try.

But it would not be in parallel like the BalancedBagging, am I correct? because estimators are built sequentially.

Also, by BalanceCascade sampler do you mean the BalanceCascade algorithm or something else? and what do you mean by BalanceEnsemble sampler?