[R-pkgs] Representation of lmer objects will change soon
This is a "heads up" to those using the lmer function for fitting
linear mixed models or generalized linear mixed models. In the next
few days Martin and I will upload a new version of the Matrix package
that employs a new representation of linear mixed models based on the
supernodal Cholesky factorization provided by Tim Davis's CHOLMOD
library of sparse matrix code.
This representation will be incompatible with the current
representation, which uses a class of blocked compressed sparse
matrices called the dBCMatrix class. If you have an lmer object
created with the current code you will need to convert it to use it
with the new package. By far the easiest way to do this is to
recreate it. (The new code is very fast.) However, we will also
offer a conversion function for those who no longer have the
information used to produce the model fit.
So the takehome message is that if you are currently using lmer then
you should save the information used to fit a model in addition to
saving the fitted model.