possible GSoC 2016 projects

[bbolker]

We're a bit late for this, so would have to decide quickly if we wanted to go ahead. Some thoughts:

• infrastructure for complex variance structures: port flexLambda up to current version, and/or proceed with lme4ord?
• implement robust standard errors via sandwich package, e.g. see #42
• implement Gauss-Hermite quadrature for models with >1 RE term/complex RE terms
• try to come up with a workable post hoc importance sampling/MCMC sampling algorithm
• implement a more flexible refit (e.g. allow refitting with new fixed-effect predictors -- i.e. anything that doesn't involve recomputing the RE structures)
• improve handling of NAs, clamping, etc. in GLMMs (e.g. step-halving PIRLS failures) (related to #188)
• ?? implement/backport efficiencies for simple (scalar-RE-only) models from DB's Julia code??
• work on nlmer ?
• a version of (g)lmer that shares big components by reference

[dmbates]

These all look interesting. I would be happy to advise (but not as the mentor) on a project to port the efficiencies for simple RE models from Julia to R. There is a proof of concept of the method and the various data structures have been blocked out. They may not translate to idiomatic R but they are a start.