1.57 - fixes and features / 1.6 candidate

- added Vehtari’s Pareto smoothed LOO comparison metric, available with
LOO(), and can be used with compare(…,func=LOO)
- map2stan gracefully accepts previous fits now and resamples without
recompiling. resample() deprecated.
- better approach to sampling inits for multiple chains
- fixed map2stan so compatible with rstan 2.8 (refresh bug)
- bug fix for link() and sim() with dordlogit models
- added WAIC method for raw stanfit objects that include log_lik
matrix. See example in ?WAIC.

Version 1.56

- a bunch of edits to man pages
- added a logit function for convenience
- uniform priors in map2stan work via constaints now and are commented
out in the model block, where they would be redundant
- added automatic positive constraints for scale paramters in
dbetabinom, dgampois, dgamma2
- fixed a subtle bug with WAIC calculations for map2stan models using
imputation
- Added quick installation section to README

1.55 - Non-centered parameterizations

- added dmvnormNC density for map2stan, automating non-centered
adaptive priors for multilevel models, using Cholesky factors
internally. This can substantially speed up sampling for many
multilevel models. See example in README.
- added a dashboard() function for checking all diagnostics on a
map2stan or Stan model fit
- added a lognormal template for map2stan: dlnorm in R, lognormal in
Stan
- trace plot grids now default to 3 columns

1.54 - Titanic Iceberg

- bugfix for map2stan imputation combined with certain kinds of varying
effects. Added a test for this.
- man pages for pairs.map2stan and plot.map2stan
- plot.map2stan (trace plots) now recognize n_cols and max_rows
arguments for arranging plot grid
- WAIC method for map2stan defaults to use all non-warmup samples
returned by Stan

1.53 - Struwen Day Edition

- man page for dlkjcorr
- added rlkjcorr function (Ben's code)
- ensemble optionally ignores link or sim
- map() now catches another corner case in linear model definitions
- map2stan plot() method for traceplots now has window and n_cols
arguments
- added divergent(), which counts number of divergent iteration in
map2stan fit. Automatically warns user.
- map2stan better handles missing values in outcome variables. Working
towards automatic prediction interface through this method.
- some file rearrangement for sanity
- precis() checks and warns about divergent iterations
- minor updates to README

1.52c: 1.52 candidate 3

- might have fixed Windows parallel chains now

Version 1.51

v1.51 - features and maintenance

- fixed some non-ASCII chars in Fieldgoals.csv
- updates to various man pages
- added some new man pages (map2stan-class, rethinking e.g.)
- removed glmm() and bglmm()
- NAMESPACE gardening continues; importing more explicitly from
dependencies now
- ensemble() now understands replace list, like link()
- some source file renaming for sanity
- fixed resample() so it uses all samples from all chains to compute
new WAIC
- map2stan(): improved handling of variable names with dots . in them;
now warns
- map2stan(): notifies user when imputing from NAs
- various plotting functions that call set_nice_margins() now respect
mfrow/mfcol settings better
- sim() now understands replace list, like link()
- removed various lingering require() calls to dependencies now in
NAMESPACE

Version 1.50

v1.50 - functionality, fixes, features

- added data(Fish)
- added pairs method for map model fits
- added pars argument for pairs methods
- map2stan works better with variables transformed with scale()
- map2stan uses more efficient Stan distributions with built-in log and
logit links
- fix for bracketed index prior declarations
- better detection of and warning about unfound variables in model
formulas
- fix to sim for map2stan models with multiple outcomes or linear models
- additions to README

Fix GP bug introduced with 1.48

In 1.48, fixing something for varying slopes models introduced a weird variable duplication bug for Gaussian processes (GPL2 distribution in map2stan). This release just fixes that.

Version 1.48

Maintenance release. Bug fixes: (1) ensemble works better with zero-weight models; (2) map models with linear models like "mu <- a" behave better.