I'm exploring a cross-test of the DirMult and a new distribution for self-weighting comps. I'm starting with a self-test of the DirMult, and initially got weird results. I think the reason is a bug in the rdirmultinom function, which should simulate a single draw from the Dirichlet for all samples from the multinomial, not a separate draw for each sample.
Do you want me to do a PR on the main or devel branch, or perhaps have some further discussion here first?
I'm exploring a cross-test of the DirMult and a new distribution for self-weighting comps. I'm starting with a self-test of the DirMult, and initially got weird results. I think the reason is a bug in the
rdirmultinom
function, which should simulate a single draw from the Dirichlet for all samples from the multinomial, not a separate draw for each sample.Do you want me to do a PR on the main or devel branch, or perhaps have some further discussion here first?