weighted adaptation target
tpapp opened this issue · comments
Stan recently switched to weighting acceptance rates by their probability. An explanation is here.
The change process was controversial, but the theory is appealing — DynamicHMC should have this as an option and test it for existing models.