search

philchalmers / mirt

Multidimensional item response theory
https://philchalmers.github.io/mirt/
203 stars 77 forks source link

add nlminb support for MHRM? #105

Closed cwliu007 closed 7 years ago

cwliu007 commented 7 years ago

Hi Phil, Will nlminb be added? I would like to check whether nlminb could perform better than optim. Thank you.

philchalmers commented 7 years ago

It's possible, but I very much doubt it would make a difference. The majority of the estimation variability comes from the stochastic MH step, where the optimiser makes very little different due to how much noise there is in the imputations. I simply included these optimizers to make a more kosher SEM/SAEM algorithm where a real M-step was being performed (as opposed to the MH-RM, which only does a single Newton-Raphson update).