Alexey Radul
Alexey Radul
Slice sampling has been solved and documented (#448).
This impinges on #470.
#453 is the plan to fix this. Marking blocked on that.
Right. In this case, the mh proposal to the 'u 0 node is not reversible, and hence computes the wrong acceptance ratio. Treating stochastically constrained nodes as potential resampling nodes...
From conversation with @luac, the following looks like a solution. It essentially amounts to making LKernels that use `amb`, but with what amounts to an ad-hoc delimited continuation structure instead...
Another use case: Block enumerating a discrete choice together with an `exactly` guarding one of its consequences. If asked first, the `exactly` will report one possible output, namely the one...
Users have occasionally come across the "Cannot make random choices downstream of a node that gets constrained during regen" error message (and its twin "Cannot make requests downstream of a...
Marking the second part of this blocked on The Plan (#453).
A place where it might matter: semi-definite covariance matrices permit degenerate Gaussian distributions whose support is not the whole space but only a subspace. Assessment would need to return float("-inf")...
This would not be difficult to add, and would be a nice introduction to the SP interface.