Tools to Visualize, Manipulate, and Summarize MCMC Output

Performs key functions for MCMC analysis using minimal code - visualizes, manipulates, and summarizes MCMC output. Functions support simple and straightforward subsetting of model parameters within the calls, and produce presentable and 'publication-ready' output. MCMC output may be derived from Bayesian model output fit with JAGS, Stan, or other MCMC samplers.


News

NEWS

0.12.2:

  • MCMCsummary fix bug where Stan input was not being sorted by parameter index

0.12.1:

  • MCMCtrace fix spacing issue for Rhat and n.eff text when Rhat = TRUE and n.eff = TRUE
  • MCMCplot add deprecation warnings for arguments

0.12.0:

  • MCMCsummary for stanfit objects (model output derived from Stan) Rhat and n_eff are calculated using the rstan package. Note that rstan calculates Rhat and n_eff values slightly differently (more conservatively) than the coda package (commonly used to summarize model output derived from JAGS).
  • MCMCplot the following argument names have been changed: orig: labels_sz, new: sz_labels orig: med_sz, new: sz_med orig: thick_sz, new: sz_thick orig: thin_sz, new: sz_thin orig: axis_text_sz, new: sz_ax_txt orig: ax_sz, new: sz_ax orig: tick_text_sz, new: sz_tick_txt orig: main_text_sz, new: sz_main_txt orig: tick_pos, new: pos_tick
  • MCMCtrace add ability to specify xlim and ylim for density plots
  • MCMCtrace add ability to specify xlab and ylab for density and trace plots
  • MCMCtrace add ability to specify title for trace and density plot
  • MCMCtrace add ability to specify line width and line type for density and prior lines on density plots
  • MCMCtrace add ability to specify color for density and priors lines on density plots
  • MCMCtrace add ability to specify size and color of text when priors specified, and the position of ticks for density and trace plots
  • MCMCtrace add ability to specify size of tick labels, axes labels, and thickness of axes
  • MCMCtrace clean up plotting of trace plots when only two plots are plotted in window
  • MCMCtrace add ability to include Rhat and number of effective samples on trace plots

0.11.3:

  • MCMCchains fix error associated with coda::mcmc and rjags objects

0.11.2:

  • MCMCplot now takes guide_axis argument. If TRUE, a second axis (x-axis if HORIZ = TRUE, y-axis if HORIZ = FALSE) is plotted to help interpret values on plot.

0.11.1:

  • MCMCchains now assigns arbitrary names to columns (parameters) when input type is matrix (along with a warning that it is doing so)
  • MCMCtrace now modifies the layout of trace plots when < 6 plots are generated
  • MCMCpstr fix bug that prevented scalars when type = 'chains'

0.11.0:

  • MCMCsummary digits argument is now NULL by default (all computed digits are returned upon default - any rounding must be explicitly specified)
  • MCMCsummary digits argument uses signif rather than round for rounding (in other words, digits specifies number of significant digits rather than number of decimal places)
  • MCMCsummary now takes round argument to round output to specified number of decimal places
  • MCMCpstr no longer has the option to restrict the number of digits output (returns all digits)
  • MCMCpstr now takes type as an argument. When type = 'summary' (default), values calculated with the func argument (default mean) are returned. When type = 'chains', posterior chains are returned while preserving the parameter structure. Posterior samples are stored in the last dimension of the array for each element of the output list. In this way vector parameters are output in matrix format, matrix parameters are output in three dimension array format (within each element of the output list - one parameter for each list element).
  • MCMCpstr now accepts output greater than length 1 from argument func. If output is greater than length 1, function output are stored in the last dimension of the array (within each element of the output list - one parameter for each list element).

0.10.4:

  • MCMCchains now takes chain_num as an argument. When specified, single posterior chains can be output for a particular parameter of interest. Useful for determining the last value in an MCMC chain for each parameter (to be used as initial values for a subsequent model run).

0.10.3:

  • All functions now return a warning for missing params inexcl and params arguments instead of an error. This means that output will be returned even when values specified do not exist. Change was made to MCMCchains code but impacts all functions.
  • Fix typos
  • Add contributors Che-Castaldo and Hardy

0.10.2:

  • MCMCtrace now takes post_zm as an argument. When post_zm = FALSE, x- and y-limits of density plots are scaled so that both the prior and posterior can be visualized on a plot (rather than zoomed on the posterior).

0.10.1:

  • MCMCpstr fix bug to institute rounding when parameters as scalars or vectors
  • MCMCtrace now takes PPO_out as an argument. When PPO_out = TRUE, the percent overlap between prior and posterior for each paras will be returned as a data.frame.
  • Add CITATION file

0.10.0:

  • Fix warning when feeding jags.parallel object to MCMCsummary
  • Fix bug that produced an error when using stanfit objects
  • Add automated tests to check package functions
  • MCMCtrace now takes open_pdf as an argument. When open_pdf = FALSE, the generated pdf will not be opened in a viewer automatically.
  • MCMCtrace now takes gvals as an argument. When simulated data are used to fit a model, the generating values used to simulate the data (true parameter values) will be plotted as vertical lines on the density plots.
  • MCMCplot ref_ovl argument now defaults to FALSE (one color is plotting for all parameter estimates)
  • Change MCMC_data (example data) so that it's smaller (only 5k iterations and two parameters)

0.9.4:

  • Fix white space issue in MCMCplot when many parameters are plotted and large fig dimensions are used
  • Fix label alignment issue in MCMCplot when horiz = FALSE and large numbers of parameters are plotted

0.9.3:

  • Fix bug that prevented parameters from being sorted when using matrix input for MCMCtrace
  • Add support for objects produced with the jagsUI package

0.9.2:

  • MCMCtrace now takes matrix input (as with the other functions). One chain is assumed when matrix input is used.

0.9.1:

  • Fix bug that produced errors when using the jags.parallel function in the R2jags package.
  • All functions - when ISB = FALSE, params argument now takes the form of regular expressions
  • Examples for MCMCtrace no longer open up external programs (pdf viewer) per CRAN policy

0.9.0:

  • MCMCpstr function now added. Function returns summary output for a specified function while preserving structure of parameters (i.e., scalar, vector, matrix, array).
  • MCMCtrace now takes a priors argument to visualize prior/posterior overlap. If specified, the prior (user specified as this information is not contained within the MCMC output) for a specified parameter is plotted on the same plot as the posterior output. Percent overlap between posterior and prior is also calculated and displayed.
  • Fix bug in MCMCchains that caused incorrect alphabetization of parameter names when output from R2jags was used.

0.8.2:

  • MCMCsummary greatly speed up calculation of Rhat values for objects with large numbers of parameters
  • MCMCchains now takes the argument mcmc.list. If specified, mcmc.list object returned rather than a matrix.

0.8.1:

  • Fix bug in MCMCsummary that displayed the same result twice when selecting only a single output parameter of interest
  • Fix bug in MCMCplot that displayed the axis label too close to tick labels when horiz = FALSE and tick labels were very long
  • MCMCsummary Rhat values always round to 2 digits
  • MCMCsummary output from func argument rounded to specified digits for rest of output
  • MCMCsummary now takes a func_name argument. If specified, column displaying output from func will be labeled with this name. If not specified, column will be labeled 'func'.
  • MCMCplot change argument x_axis_text_sz and x_tick_text_sz to axis_text_sz and tick_text_sz respectively

0.8.0:

  • Specification of parameters of interest now works slightly differently. The argument ISB (Ignore Square Bracket) has now been added. By default ISB = TRUE - params and excl match exactly to parameter names by default (ignoring square brackets). When ISB = FALSE, square brackets will not be ignored, and will match on partial names (as when using grep). This applies to all functions.
  • MCMCsummary now takes a func argument. If a function is specified, it will be evaluated for all specified parameters and specified in the MCMCsummary output.
  • MCMCsummary speed greatly increased. Parameters of interest are now sorted before calculations are made. Rhat values are no longer masked, but rather not calculated when Rhat = FALSE. These changes result in dramatic speed ups for large objects.
  • MCMCsummary bug fixed that caused function to fail when only one chain was run
  • MCMCsummary standard deviation added to summary output for each parameter.
  • MCMCsummary number of effective samples added to summary output for each parameter. Default is n.eff = FALSE (metric will not be calculated or displayed).
  • MCMCtrace default is now to write trace plots to pdf. Default number of iterations changed to 5000 from 2000.
  • MCMCplot y-axis labels now vertical when horiz = FALSE to improve readability.
  • MCMCplot bug that resulted in poor plot dimension choices in some circumstances now fixed.
  • MCMC_data now contains three chains with 6000 iterations each.
  • Error message now added about functions not taking objects produced from jags.samples function in the coda package. coda.samples should be used instead.

0.7.1:

  • Fix bug in MCMCplot which incorrectly shaded parameter estimates when plotted vertically
  • MCMCsummary now displays estimates for deviance with MCMC output fits with R2jags

0.7.0:

  • Fix bug in MCMCsummary to do with Cholesky decomposition and calculating Rhat.
  • Speed up processing of MCMCsummary for certain object types.
  • Fix minor documentation errors in help files for several functions.
  • MCMCplot labels now start at top and go down (more intuitive).
  • MCMCtrace now plots only the last 2000 iterations of the posterior chains by default. The argument is now number of iterations to be plotted, rather than the starting iteration to plot. As such, MCMCtrace argument iter_st (start iteration) changed to iter (number of iterations from end).
  • Add horiz argument to MCMCplot - caterpillar plots can now be plotted to run vertically rather than horizontally. Parameters are plotted left to right when plotted vertically.
  • Remove extended lines on axes for MCMCplot - axes lines only goes to the end of ticks now.

0.6.3:

  • Initial release

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("MCMCvis")

0.13.5 by Casey Youngflesh, 2 days ago


http://github.com/caseyyoungflesh/MCMCvis


Report a bug at http://github.com/caseyyoungflesh/MCMCvis/issues


Browse source code at https://github.com/cran/MCMCvis


Authors: Casey Youngflesh [aut, cre] , Christian Che-Castaldo [aut] , Tyler Hardy [ctb]


Documentation:   PDF Manual  


Task views: Bayesian Inference


GPL-3 license


Imports coda, rstan, methods, overlapping

Suggests knitr, rmarkdown, testthat


See at CRAN