It seems like it might be useful to be able to control the number of MH steps that are taken between each HMC iteration. Doing a long HMC simulation where all the continuous variables may move a long way and then only flipping a single discrete choice afterwards seems kinda unbalanced.
If this is useful, we could probably implement it by exposing our kernel combinators sequence and repeat.
we should expose these kernel combinators anyhow... eg have {kernel: {sequence: [{kernel: MH ...},{kernel: HMC ...}]. there are other useful things to do with sequence and mixture kernels...
From a comment on #420:
If this is useful, we could probably implement it by exposing our kernel combinators
sequence
andrepeat
.