Kira Selby
Kira Selby
Just a comment that I have tested this with a MCMC sampler I am using. When I use the existing posterior update, the sampler does not converge, and gives nonsensical...
After looking at this further, it seems like the T that is intended to be stored in the Normal-Wishart object here is the inverse scale matrix for the Wishart, rather...
I believe I updated the PR to reflect the inverse scale matrix parametrization. It would be good to have someone look it over though to make sure I didn't miss...
After checking again, the pdf function actually did need to be changed for this parametrization. I also fixed a syntax error in logpdf, and commented out a line which called...
Apologies for making so many commits. I have now gone through and thoroughly tested this and am confident it is now right.
I am working on writing something up, I will hopefully post it later today.
To be honest I am still not entirely sure how best to approach the problem of how to parametrize this - but no matter how we decide it is best...
I'll try to get to this today and look over everything again. Sorry I've been dragging my heels, I've just been caught up in other work.
Go for it!
After taking another look, I believe the alternate parameterization we were considering here is actually the parameterization for the Normal INVERSE Wishart. I returned to the original parameterization, and added...