2014-02-05 16:56 Vincent Dorie
* Brought to parity with the latest lme4 release (1.1-6).
* Bug fixes:
quadratic term for normal priors calculated incorrectly when using non-zero correlations
glmms could fail to apply full penalty term for mixed use of covariance priors
* New features:
t priors for fixed effects
custom priors for ranef covariances
2013-09-13 03:30 Vincent Dorie
* Complete re-write for lme4 version 1.0.
2013-08-12 10:25 Vincent Dorie
* Minor addition to testing suite set of functions. Upgraded to lme4 0.999999-2.
* Changed dependency to lme4.0.
2012-11-13 10:25 Vincent Dorie
* Common scale is now optional for random effect covariance, specified with the "common.scale" parameter and taking values TRUE/FALSE.
* Covariance priors can now all be improper.
* Common scale prior can now be inverse gamma on sd, or gamma.
* Reworked how the penalized weighted residual sum of squares is calculated, as it may not make sense depending on what priors are used.
* Brought to parity with the latest version of lme4, 0.999999 or so.
* Eliminated a false zero problem wherein the optimizer takes an initial step to the boundary and get stuck due to a flat likelihood.
2012-10-13 10:11 Vincent Dorie
* blme now ignores the common scale when penalizing the random effect covariance. In previous versions it multiplied it in but didn't take the prior into account when profiling it out - in essence failing to correspond to a probabilistic model. The next version will allow the prior to be on the scale-free covariance, or on the real-world covariance matrix. For now, just the scale-free version. ** This means the fitted models will be different than previous versions. **
* Reduced the default shape for random-effect covariance posteriors on the variance scale so that the polynomial term has an exponent of 1 in the univariate case. The multivariate now has df of dim + 1 to correspond.
* Fixed a few typos in the documentation.
* Cleaned up imports from NAMESPACE file.
* Replaced a .Internal call.
2012-03-09 13:01 Vincent Dorie
* Fixed off-by one bug on termination of the optimization loop which caused gradient evaluation steps to be reported a the maximum.
* REML optimization for sigma with a fixef prior was slightly off.
* sigma now accepts inverse gamma priors (improper, as well).
* Internal version of sim().
2011-12-15 17:40 Vincent Dorie
* Prevented summary from printing misc. lines when only one covariance prior is implemented. Gaussian fixed effects priors summarize in terms of their covariance correctly when the covariance is diagonal or a scalar times the identity.
2011-11-15 17:03 Vincent Dorie
* Fixed a bug related to the scaling of priors to the data in the presence of interactions.
* Added point priors on the common scale, enabling meta-analyses.
2010-12-19 19:27 Vincent Dorie
* Initial version.