Closed OkonSamuel closed 1 year ago
I've tried to update it, but the AUC evaluations fail because that measure only supports probabilistic
types. Maybe we should stick to probabilistic
for the SIRUS models? They all depend on randomness inside the forests, so they are not fully deterministic (they are when picking a rng, but not without).
I'm confused as to why you need auc here? Is it something used in fitting the model? Can you shed more light. Maybe we can wrap the output in a simple distribution wrapper. @ablaom what do you think?
Oh wait. I have accidentally updated a classifier. My bad! I'll try again to fix this!
@rikhuijzer The
StableRulesRegressor
andStableForestRegressor
models must subtypeDeterministic
notProbabilistic
as callingpredict
on these model types doesn't return a probabilistic distribution as shown in the code below.The relevant code lines are https://github.com/rikhuijzer/SIRUS.jl/blob/main/src/mlj.jl#L69 https://github.com/rikhuijzer/SIRUS.jl/blob/main/src/mlj.jl#L78