Is there a way in fastai library to combine the results of multiple CNN architectures e.g. ResNet50, ResNext101_64 etc. by majority voting or averaging out etc.?
I think it should improve the results as ensembling generally does in case of Bagging (Random Forest).
I do not think there is an automated way of doing this. One way to go about this would be to put the results of all the model in a numpy array / pandas dataframe where each row contains outputs of all the models that you trained for a specific example. You could then apply a function to the rows to combine the results, either via majority vote, average, weighted average, etc.