Is your feature request related to a problem? Please describe.
The new subject-specific epsilon values mean that adaptation may finish with iterations equal to almost the minimal number of unique samples (currently 20). This is a bit low, and therefore we should increase the minimal number of unique iterations to be larger, say 100.
An additional improvement may be that if the proposal distribution creation fails that we should increase the iterations by a smaller amount.
Is your feature request related to a problem? Please describe. The new subject-specific epsilon values mean that adaptation may finish with iterations equal to almost the minimal number of unique samples (currently 20). This is a bit low, and therefore we should increase the minimal number of unique iterations to be larger, say 100.
An additional improvement may be that if the proposal distribution creation fails that we should increase the iterations by a smaller amount.