Given a model, for each prediction (after summing all the individual rules) the classes should sum to 1. Many other algorithms do this, but SIRUS.jl not. For example, it is possible with SIRUS.jl to get an output where the end prediction for some observation is [0.001, 0.005] which in total is only 0.006 and not roughly 1.0. Scaling these numbers closer to 1 would make model interpretation easier.
Thanks to Matti Ruuskanen @Begia for this observation
Given a model, for each prediction (after summing all the individual rules) the classes should sum to 1. Many other algorithms do this, but SIRUS.jl not. For example, it is possible with SIRUS.jl to get an output where the end prediction for some observation is
[0.001, 0.005]which in total is only 0.006 and not roughly 1.0. Scaling these numbers closer to 1 would make model interpretation easier.Thanks to Matti Ruuskanen @Begia for this observation