Updating Stan Models to be more performant

[SteveBronder]

Looking over the recent paper here I noticed that a number of the models they outperform stan on need updated and are not really made for performance testing. I'll list out a few notes on these as I go

earn_height.stan:

• seperate beta into two reals alpha and beta

hmm_example:

• @bbales2 @charlesm93 could all of these hmm examples use the new hmm stuff?

kidscore_interaction

• Make data into matrix, intercept as real, and beta's as vector to use normal_id_glm
• I think this applies to all of the kid models

log10earn_height:

• Use vectorized version of log10
• @t4c1 should we have a specialized version of normal_id_glm() that works for vector ~ normal_id_glm(vector, real, real , real)?

logearn_interaction

• For all these I think they can be made ot use normal_id_glm()
• I think for all these models the tranformed data should be written inline like

vector[N]  log_earn = log(earn);
vector[N]  inter    = height .* male;

logmesquite_logva

• normal_glm

mesquite.stan

• normal_glm

pretty much all of these can use normal_id_glm

arK

• Actually idk but it looks like there is a way to make this more performant.

[t4c1]

@t4c1 should we have a specialized version of normal_id_glm() that works for vector ~ normal_id_glm(vector, real, real , real)?

I doubt this would bring significant speedup over the code in the model. The performance bottleneck is computing the vector related stuff, so the differences in a few scalar operations is negligible.

[MansMeg]

That seems like good improvements. Could you open a PR with these changes?

[SteveBronder]

srry I don't have time I'm just posting this for someone who wants to contribute