0.19-1 Martyn Plummer
- gelman.plot() now avoids superfluous calculation of multivariate
diagnostic. Thanks to Gert van Valkenhoef.
- Fixed various issues with the histogram produced by densplot()
for discrete-valued distributions. Thanks to Robert Goudie.
- Obsolete INDEX file removed
- Dimnames bug in geweke.plot fixed. Thanks to Jiri Moravec
0.18-1 Martyn Plummer
- Arni Magnusson is fully credited as coauthor
- Fixed bug in "varnames<-", which failed when the mcmc object was
not a matrix. Thanks to Evangelos Evangelou and Pavel Krivitsky.
- NAMESPACE file now includes imports of functions from base packages.
- codamenu no longer gives option to save mcmc object on exit.
0.17-2 Martyn Plummer
- Avoid redundant copies when subsetting mcmc.list objects
0.17-1 Martyn Plummer
- Moved lattice package from Depends to Imports in DESCRIPTION
file. This means that you must load the lattice package to use
any of the lattice methods provided for mcmc objects.
- Russell Almond is fully credited as co-author
- Moved AUTHORS file into inst/ sub-directory
- Fixed bug that stopped trellis graphics from working with vector
mcmc objects (thinned chains). Thanks to Chris Andrews.
- Added additional arguments (...) to as.mcmc()
- coda.options() no longer writes to the global environment
0.16-1
- Fixed main title bug in densplot
- Substitute spectrum0.ar for spectrum0 internally
- densplot parameters fixed when drawing multiple plots
0.15-3
- Correct x axis labelling in xyplot. Thanks to Pavel Krivitsky.
0.15-2
- Allow default plotting parameters to be overridden in densplot and traceplot.
- Stop overplotting of user-supplied title in densplot.
- Fix xyplot for univariate chains. Thanks to Pavel Krivitsky.
0.15-1
- Use Authors@R field in DESCRIPTION
- Fixe geweke.diag for long mcmc samples. Thanks to Philip Johnson
0.14-7
- In gelman.diag, the multivariate potential scale reduction factor
is now optional and can be turned off with the argument multivariate=FALSE.
0.14-6
- Removed deprecated top-level file COPYING
0.14-5
- Replaced defunct restart() function in codamenu with with try()
- Changed as.ts.mcmc to be a method for generic function as.ts
0.14-4
- Arguments of gelman.plot are now consistent with gelman.diag.
0.14-3
- Fixed documentation errors in gelman.diag (Thanks to Peng Yu)
- Added CITATION file
0.14-2
- Fix documentation bugs (Thanks to Kurt Hornik)
0.14-1
- traceplot and plot.mcmc no longer plot a smooth line by default,
although it can be added using smooth=TRUE
- new head and tail methods for mcmc and mcmc.list objects
0.13-1
- as.mcmc.list generic and default method are exported in namespace
0.12-1
- Added further changes for S-PLUS from Dawn Woodward.
0.11-3
- Fixed documentation on subsetting methods for mcmc and mcmc.list objects
0.11-2
- The codamenu function has been modified so that it no longer works with
coda.dat and work.dat in the global environment, solving global binding
issues.
- The sample size test at the beginning of codamenu has been corrected
for the case of multiple chains.
Thanks to Luwis Diya and Pablo G Goicoechea for these bug reports
- Syntax errors fixed in help pages.
- Description of Gelman and Rubin diagnostic corrected
0.11-1
- Added changes from Dawn Woodward for S-PLUS.
- Fixed global binding problems
0.10-7
- Documentation clarification for bugs2jags()
- Made linearity test in codamenu less stringent.
Thanks to Bengt Sennblad.
- Fixed rejectionRate() for single-variable chains.
Thanks to Rob Scharpf.
0.10-6
- Fixed HPDinterval() for univariate chains
Thanks to David LeBlond
- Fixed optional parameters in as.data.frame method for mcmc objects
0.10-5
- Registered default method for as.mcmc in NAMESPACE
Thanks to Christian Grose
0.10-4
- fixed eternal loop in codamenu() when saving plot as PostScript.
Thanks to Bengt Sennblad.
0.12-3
- added 'data' in lattice generics for compatibility
0.10-2
- Import generics from lattice
- In plotting functions, "ask" now defaults to dev.interactive(),
and not the default value in par(), as introduced in 0.9-2
0.10-1
- Added "mcmc" and "mcmc.list" methods for several lattice functions
(xyplot, qqmath, densityplot and acfplot (with generic here as
well)). There's a levelplot method too, which is currently
experimental. These are Trellis analogs of existing coda functions,
and may in future replace them.
0.9-5
- Fixed bug in summary.mcmc (safespec0 misspelled) which affected
univariate chains.
0.9-4
- Fixed documentation errors.
0.9-3
- Added date stamp check in read.openbugs to ensure files were created
at the same time.
0.9-2
- [RGA] Added an autoburnin flag (default) true to gelman.diag to suppress
automatic windowing for burn in (so I can do it manually).
- [RGA] Fixed a problem where summary.mcmc.list would not give correct
pooled standard errors.
- [RGA] Fixed propagation of standard par and titles to plot.mcmc and
plot.mcmc.list. Also fixed so that ask will default to value in
par().
- [RGA] Fixed a problem where autocorr would apply thinning twice to
mcmc.list objects.
- [RGA] Changed effectiveSize for mcmc.list to sum across all chains.
Original behavior can be recovered by using lapply(x,effectiveSize)
- [RGA] Added autocorr.diag function.
- [RGA] Patched summary.mcmc and summary.mcmc.list so it would give
standard error of NA when spectrum0 blows ups.
- [RGA] Added a rejectionRate method.
- [RGA] Added a batch Standard Error Function
0.9-1
- spectrum0 function now has default max.length argument of 200.
This means that the output will be batched to a length between
100 and 200 before fitting the glm to the spectrogram. This
should improve robustness for chains with high autocorrelation,
or markedly non-gaussian distributions.
- The read.bugs() function has been removed; read.coda() has been
modified to allow specification of both output and index files;
read.bugs.interactive() has been modified in the same way;
read.openbugs() is a new wrapper function around read.coda() for
OpenBUGS output.
0.8-3
- Fixed documentation errors in coda.menu.Rd and linepost.Rd
0.8-2
- Added generic function thin to list of exported functions
0.8-1
Continuing problems with namespace:
- Addition of namespace requires new version number for coda,
as saved workspaces are not backwards compatible.
- Ensured that mcmc attributes are not assumed for objects returned by
as.matrix.mcmc, as this now no longer returns an mcmc object.
This occurred in, for example, effectiveSize (which returned NA)
and gelman.diag (which dropped variable names).
- Changed "[.mcmc.list" and "[.mcmc" so that they return an
mcmc.list/mcmc object respectively when subsetting columns.
plot.mcmc() and plot.mcmc.list() now work again when there is
only one variable.
- Imported the required time series generics from package stats.
Failure to do this may result in a saved workspace that cannot be reloaded.
Other problems
- Removed S compatibility (statements conditional on is.R and wrapper
function coda.global.assign). Note that S compatibility never worked
at all, and I now have no intention of supporting it.
- Fixed (old) bug in "[.mcmc" which made column subsets of mcmc objects
return invisibly.
- Changed the plotting functions so it is no longer necessary to press
return to see the first page of plots.
0.7-3 (changes done by Andrew Martin )
- Added NAMESPACE
- export only functions that I think should have been exported (based on
the documentation and common sense).
- Registered all S3 methods.
- Fixed documentation for non-exported functions.
- Patched the "mcmc" function to deal with really big thinning intervals.
- Patched the "as.matrix.mcmc" function so it really returns matrices.
- Fixed plot method, by having plot.mcmc pass an mcmc object
rather than a matrix, and by fixing the [.mcmc.list method so it
returns an mcmc.list not a matrix.
- Fixed "varnames<-" which was broken when the as.matrix method was
fixed.
- Fixed documentation mcmc.convert.Rd such that the usage is consistent
with S3 class definition (this fixed an error thrown in the QC tools).
0.7-2
- The spectrum0() function now returns zero when it is given a constant
vector. summary.mcmc() (which calls spectrum0) now works correctly.
0.7
- Modified to run on R 1.9.0 with new organization of base library
0.6-2
- Fixed documentation bug in raftery.diag
0.6-1
- spectrum0.ar no longer crashes when the chain is a linear function
of the iteration number
- codamenu automatically drops variables that are linear functions
of the iteration number
- read.bugs renamed to read.coda. read.bugs exists as an alias.
- initial support for JAGS: bugs2jags function converts WinBUGS
data to R dump format used by JAGS.
- added cumuplot function (not yet in codamenu system)
0.5-14
- Fixed reporting of sample size in densplot(), for variables that are
in the range [0,1] or [0,Inf). Thanks to Roy Levy.
0.5-13
- mcmc() now works with data frames (provided that they contain only
numeric values).
0.5-12
- Documentation errors in coda.options.Rd and nchain.Rd fixed.
Thanks to Kurt Hornik.
0.5-11
- Fixed bug in mcmc() function that causes problems in Geweke diagnostic
- Fixed geweke.plot so that it works with samples running from iteration
N to 2N. These were previously mis-diagnosed as being too short.
Thanks to Vasco Leemans for finding both bugs.
- In autocorr.plot, the `ask' argument was not used. Thanks to
Dennis A Wolf.
0.5-10
- Eliminated use of "=" for assignment operator. This is a syntax error
for R < 1.4.0
0.5-9
- Further documentation bugs removed
0.5-8
- Removed further documentation bugs found by "R CMD check coda" using
R-1.4 (pre-release)
0.5-7
- Removed obsolete line.doc and line.old.doc files from data directory.
- New spectrum0.ar provides model-based estimate of spectral density
at frequency zero.
- New effectiveSize diagnostic gives effective sample size.
- codamenu includes automatic check on effective sample size.
0.5-6
- Removed further documentation bugs found by "R CMD check coda" using
R-1.3 (pre-release)
0.5-5
- Ironed out last warnings generated by "R CMD check coda"
0.5-4
- Provided documentation for all functions and datasets.
(Thanks to Kurt Hornik for the prompting)
0.5-3
- Fixed bug in example for mcmc.list. Row subsetting no longer preserves
mcmc objects (Thanks to Kurt Hornik).
0.5-2
- Fixed bug in gelman.transform which did not work for univariate chains
(Thanks to Mark A. Beaumont)
- Fixed confidence limits in geweke.plot
(Thanks to Mark A. Beaumont)
- Allow user to set ylim in densplot (Thanks to Niels Peter Baadsgaard)
0.5-1
- Replaced time series functions with functions from "ts" library
(R-base >= 0.65.0)
- Removed calls to Version() (deprecated). Use is.R() instead.
- Added new function read.yesno
- Source files maintained using ESS
- Allowed restart() in codamenu.options.plot.kernel
- Simplified print.coda.options
- Removed "onepage" option in coda.options() (Subsumed in user.layout)
- Removed "mrows" and "mcols" options in coda.options (use par instead)
Changes to Geweke's diagnostic
- Uses new function spectrum0() to estimate spectral density at zero
- Gelman-Rubin-Brooks plot never discards more than half the chain
to preserve necessary asymptotic conditions.
Changes to Gelman and Rubin's diagnostic
- Multivariate psrf added.
- Documentation for Gelman-Rubin-Brooks plot update to give
clearer motivation.
Changes to Heidelberger and Welch's diagnostic
- Simplified formula for Cramer-von Mises statistic
- Using new function spectrum0() to estimate spectral density at zero
- Can set p-value threshold for passing convergence test.
- p-value is printed in output, using new function pcramer().
- Prints starting iteration of truncated chain instead of number
of iterations to discard.
0.4-7
Fixed bug in read.bugs.interactive() leading to failure when
user enters both ".ind" and ".out" names (Thanks to John Logsdon).
0.4-6
Archive 0.4-5 was incorrectly compressed with "compress" instead
of "gzip". Corrected by Friedrich Leisch.
0.4-5
Bug fixes
- densplot failed with show.obs=TRUE when scale was "positive" or
"proportion".
- as.matrix.mcmc failed to preserve start, end and thin.
- codamenu did not tidy up on exit.
0.4-4
Started S3 compatibility
Fixed bug which caused options menus to crash
Fixed legend bug in gelman.plot
Confirmed that these bugs are fixed:
* densplot "missing" and "scale" bugs (Thanks to Greg Warnes)
* autocorr "improper time series parameters" bug ("acf"
function rewritten by Paul Gilbert)
* integer overflow bug in raftery.diag (Thanks to Morten
Frydenberg)
* read.bugs.interactive will search for the files it needs
and print their names.
codamenu now assigns default variable and chain names to
data when these are NULL.
0.4-3
Fixed help page errors pointed out by Brian Ripley.
Fixed coda.credits
0.4-2
Whoops. 0.4-1 was a mistake.
0.4-1
updated manual pages
new class mcmc.list added to deal with multiple chains. ugrade.mcmc
function introduced to deal with old mcmc objects.
plot functions changed to use the "ask" parameter instead of
"pause" functions.
spec.pgram now handles matrix time-series.
acf function now calculates cross-correlations. Thanks to Paul Gilbert.
codamenu functions now use title argument in "menu"
corrected spelling mistakes in help pages
changed instances of "T" and "F" to "TRUE" and "FALSE" (R coding standards)
densplot now recognizes discrete distributions and prints histogram ...
... also prints histogram if IQR=0 (large mass on one point)
fixed bug in mcmc which allowed non-integer thinning intervals
Fixed manual pages with bad use of "alias" command.
0.3-4
Fixed bug in "tspar<-" which breaks much of the code in R-0.62
0.3-3
Package was in obsolete format. Corrected by Fritz.
** Pre-release changes for R version
Created class "mcmc" with associated constructor and extractor
functions as well as plot, print and summary methods.
Modified diagnostics so they all work on objects of class "mcmc"
and can be called directly by the user. All diagnostics return
objects with associated print methods. Renamed some functions
and arguments for ease of use.
Changed the menu driven interface - now called by the function
"codamenu" - to avoid recursive calling of menu functions.
Put frequently used code inside utility functions.
Got rid of functions written by Mathsoft
Wrote drop-in replacements for some time series functions which
are found in S-PLUS but not R.
The logfile facility has been removed. Sorry.
Changed license terms to GPL.