Echidna now sets variances in a diag() structure which approach zero to exactly zero which is equivalent to dropping the associated effects from the model. Be wary in case this new feature doesn’t work properly in your particular model.
Several issues with fitting Factor Analytic models have been identified and resolved.
When fitting a variance structure which is the interaction of a genetic relationship matrix and a factor (such as Trait or Site), the nrm(Site).(rrk(Site) + diag(Site) ) specification usually runs faster than the model equivalent alternatives.
Have adjusted updates of Variance (P) parameters; faster to boundary.
Add an AI Singularity notice
Set gamma to zero when at boundary (simple random factor)
Set gamma to zero when at boundary (diag())
Bug in PREDICT when using augmented data and !PRESENT has been fixed
Revise AI update checks for US; Single step EM
Adjust !CYCLE processing.
Improve initialising of rr() when starting from 1 less factor.
Fix several bugs in rr2 models; score, effects and hence residuals
