NEWS.md
pkgdown stuff - not much more needs to be said
here.pkgdown
website.get_p() became get_params()
match_p() became match_params()
base_p() became drop_index()
p argument now
accept a params argument (match_params(),
diag_plots() post_subset(),
post_summ(), post_remove(),
vcov_decomp(), etc.) (#14)p_summ has been replaced with probs in
post_summ() (#25)match_params and get_params now both take
a type argument, and both accept only values of
"base_only" or "base_index" (#20 - some other
function and arguments changed here as well).density_plot() and trace_plot() previously
accepted a p_one argument – this is now param
(distinction between multiple parameters and one parameter); but most
users won’t notice this as these are non-exported functions.ins_regex_bracket(),
rm_regex_bracket(), ins_regex_lock(), and
rm_regex_lock() now take the params argument
rather than the string argument, but most users won’t
notice this.matrix2mcmclist() was deleted and now
post_convert() does its job (#23).par() after completing diag_plots()
(#17)StatonMisc dependencies (#16)post_thin()
thin_percent is not a
perfect multiple of the number of iterations per chain - this causes the
thinning to be “quasi-evenly” spaced. As this function is only ever used
in code development and not final products, this should not be an issue
(#13).thin_percent to be keep_percent,
and implemented throughout.keep_iters argument – now users can choose to
keep either a fixed number of samples (again, may be quasi-equal to
keep_iters, due to rounding errors) or a fixed percent of
samples.by_chain argument to post_summ()
(#10)
post_summ(..., by_chain = TRUE). For now, Rhat, ess, and
mcse will be ignored if by_chain = TRUE.post_convert() functionpostpack requires (for the most part) that MCMC samples
be stored in mcmc.list objects. But not all R packages that
perform MCMC return samples immediately in this format. This new
function may help users get their samples into mcmc.list
format, potentially broadening the accessibility of the functions in
this package.density_plot() and trace_plot()
from exported listpost_remove functionpost_subset, except that it prompts the user to
verify that they wish to exclude the match nodes and returns everything
that does not match the p argument.post_summ (#11)auto_escape option (#5)mcmc.lists (#7,
#9)
post_subset() and post_thin() previously
used coda::as.mcmc.list() to convert matrices to mcmc.list
objects, however, if the matrix contained only one node, this would
result in the mcmc.list no longer having a name for that node. Because
most of the postpack functions rely on node names, this caused issues
(e.g., diag_plots(), post_bind()).matrix2mcmclist to perform
this task, which should fix these issues.Bug fix when calculating Rhat
When calculating Rhat using post_summ(..., Rhat = T)
(which wraps coda::gelman.diag) on an object containing
only one node, an error would be thrown:
Error in data[1:nobs, , drop = FALSE] : incorrect number of dimensionsWhich I tracked down to be a result of the autoburnin
argument being true by default. Being able to perform the calculations
for only one node was required for the
diag_plots(..., show_diags = T) update. So I have forced
this argument to be FALSE in its usage within
post_summ().
vcov_decomp(): it now
internally uses id_mat()
post_bind() now allows one of the objects to be a
matrix of derived quantities, but still requires that it has the same
number of samples as the object it is being binded to.diags_plots(..., show_diags) argument to display Rhat and
ESS on the density plotmatch_p now uses
StatonMisc::list_out to print the unique node names, which
makes its printing much cleaner (#2)warn argument from all functions that call
match_p
prettify argument to
post_summ() was removedpost_dim(..., kind) was included
in other functions including:
post_bindpost_thinvcov_decompvcov_decomp has improved error messages and relies on
StatonMisc::progress_updater to print the progress of the
calculationpost_dim improved to rely on
StatonMisc::list_out
id_mat functionpost_dim() now accepts a kind argument,
which represents certain element(s) from the returned vector.vcov_decomp() function
mcmc.list object with these nodesdiag_plots (#6)
diag_plots(..., layout = "2x1") and
diag_plots(..., layout = "4x2") added, and the
"auto" option will now consider these when selecting the
best option.write_model(): I essentially duplicated
R2OpenBUGS::write.model because it was the only function I
ever used from that package now all my BUGS models are fitted with JAGS.
post_dim was counting the chain and iters
columns as nodes (#1){codaTools} to
{postpack}. This is a better name, because it is not just
used for convergence diagnostics.get_nodes and filter_post:
better error messagesformat = "matrix" to matrix = T/F
argumentmatch_p function - now all
subsetting is handled by this one functionfilter_post became post_subset
ins_regex_lock,
rm_regex_lock, and base_p - all intended to
improve the behavior of diag_plots
StatonMisc functions, they are now
imported rather than requiring the whole package at the start of a
function that needs themsumm_post became post_summ
native_format became array_format
get_nodes became get_p
bind_post became post_bind
thin_post became post_thin
post_dim
diag_plots through the
my_title functionthin_post, native_format
bind_post: combines two mcmc.list
objectsthin_post: thins an mcmc.list object at
regularly-spaced intervals from each chain