In our current implementation, only one detector instance is updated each time step (the one corresponding to the true class of a given sample).
We should instead try updating all detector instances connected to the base classifier instances that got updated, as is the case in the River ARF implementation. This will require running experiments with the two versions to compare performance.
In our current implementation, only one detector instance is updated each time step (the one corresponding to the true class of a given sample).
We should instead try updating all detector instances connected to the base classifier instances that got updated, as is the case in the River ARF implementation. This will require running experiments with the two versions to compare performance.