Is it possible for an ensemble classifier to return bimodal vote?

Question

Knowing fully well that Majority and Plurality voting of ensemble classifiers for prediction of a class label returns the modal prediction by each base classifier, if there's an ensemble of about 4 classifiers and above, is it possible to get a bimodal or trimodal votes? If possible, what will the ensemble return as the predicted class label?

Answer 1

I think it depends on the classifiers that you are trying to ensemble. If the number of possible labels/classes of each classifier is the same, then 'reducing' a 4 classifiers or above to 2-3 votes can be done using some function/layer. A common example of a layer you can use to do that is a fully connected neural network layer, which is readily available in many popular libraries (eg Tensorflow, Pytorch, Caffe...)

In case the types of outputs from each classifier is different, you may need to 'normalize' the classifiers' outputs so that they have the same types and the same ranges of values. Then these normalized outputs can be used to produce bimodal or trimodal votes, similar to what was explained above.