Matthew Stephens
Matthew Stephens
that's good. A few things to resolve: - you wrote a replacement to my "gradient" function that seems to be a fix to a divide by zero problem. I guess...
for autoselect, maybe we could use the cdf instead of the moments? I only used moments in the normal case as a shorthand for making sure the grid spanned "plausible"...
ash and ash.workhorse are the same - only difference is how many options are documented. i just thought there were too many options for your average user for what is...
maybe we should introduce ash_pois , ash_binom etc for ash with other (non-normal and non-t) likelihoods? I think this could have multiple advantages. For example it seems maybe the defaults...
thanks @pcarbo . I agree with much of what you say. I'm happy to combine the help for ash and ash.workhorse. I think one benefit of having `ash.pois`etc would be...
I chose .pois to be consistent with ash.workhorse and smash.pois I generally try to use _ but here it seems when you have "variants" of a procedure the . notation...
@mengyin @jhsiao999 you could try reducing rtol as it suggests (in control =list(rtol=...))
@mengyin I'm not sure 0.1 is actually small (or what the default is) it was just an example. Id try 1e-8 Which method gives the higher look likelihood and how...
Sorry, that was loglikelihood