I’ve got a situation where I’m trying to select a set of “discriminators” from among a large number of things… so I want a vector of either -1,0,1 which when dotted with a vector of 0,1 (absent,present) gives a “score” that i can use like a logit.
Since this is a discrete distribution I’m doing MH() sampling, and I want to do a very large number of samples, like 1M but I want to keep only every 100th sample or something. In the end I’d have 10k samples but hopefully much less autocorrelated than if I had just done 10k samples.
sample(model, MH(),1_000_000; skip=100) #or something
How can I make Turing keep only every 100th sample?
There doesn’t seem to be anything in the docs?
EDIT: by reading the code I see there’s a “thin” option? is this the right thing? If so, is there somewhere this is described in the docs? If not, where should it go, I’ll do a PR.
sam = sample(fmmodel,MH(),50000; discard_initial=2000,thinning=20)
For example, looks like the options I’m looking for are “thinning” and “discard_initial”