I would like to sample from a multivariate PyTorch distribution.
This is a complex learned distribution and not a standard distribution (specifically, a TransformedDistribution created with pyro).
Doing this with ALL missing variable values is simple: I just use .sample
However, I have some known values and some not, and I would like to fill in the values for the missing variables (impute) by sampling from the distribution, with given constraints on certain variables. In other words, I would like to condition my sampling on some known values.
Of course, I can't just create a new sample from the distribution and then change the known variables to the known answers, because I need to use the information I already have to inform the sampling and constrain my sample space.
This seems like it should be straightforward; yet, I do not know how to do it. Any help would be appreciated, thanks!
