For some problems we expect that the proposal distribution should be recursively defined. We've analogised this as the proposal distribution being described as a N-sided dice where one or more of the sides/results can be "do and roll again".
We would like to be able to describe these once such that both sampling from and evaluating the proposal distribution can be done from one "description". This will most likely require some lazy evaluation of the recursion.
For some problems we expect that the proposal distribution should be recursively defined. We've analogised this as the proposal distribution being described as a N-sided dice where one or more of the sides/results can be "do and roll again".
We would like to be able to describe these once such that both sampling from and evaluating the proposal distribution can be done from one "description". This will most likely require some lazy evaluation of the recursion.