Professuren für Statistik und Ökonometrie
Schrift vergrößern Schrift verkleinernBarrierefreie Version
Suche | English
bayesicon





BayesX




Frequently Asked Questions



    This page contains a list of frequently asked questions concerning the usage of and difficulties with BayesX. If your question is not answered here, you can contact us via bayesx@uni-goettingen.de. If you want to report a bug, please include any files that are needed to reproduce the bug and describe the problem as precisely as possible.



  • I can not start the GUI version of BayesX under Windows
    Running BayesX in the GUI version requires a java installation (64 bit, version>=1.8). In addition, java has to be included in the path to be able to run the GUI (see for example here on how to adjust the path variable). You can check whether java is available and which version you have installed by typing java -version in the command line prompt. In general, it is also recommended to run BayesX in administrator mode.

  • Does BayesX support spaces in file names?
    Unfortunately not always. Altough spaces in file names work well in some situations, they should in general be avoided. In particular, an installation directory without spaces should be chosen.

  • Does BayesX support the specification of relative paths?
    Again, unfortunetaly not always and not on all operating systems.

  • How do I obtain a map in boundary or graph format?
    Whenever the coordinate (or polygon) information on geographical regions is available, it is (at least in principle) possible to construct a boundary file by following the description in Chapter 5 of the reference manual. The BayesX R package contains some functionality that allows to convert shapefiles into boundary files and some additional functionality for modifying existing boundary and graph files. If you have successfully converted your geographical information to a boundary file, we would be happy to make it available in the maps area to other BayesX users.

  • Does BayesX support non-numerical variable types (such as character variables)?
    Unfortunately not. All variables will have to be converted to numerical values before they can be read into BayesX.

  • Is there a connection to R? Can I call BayesX from R?
    Yes, the R-packages R2BayesX (available from here) and BayesR (available from here) provide easy and convenient access to the BayesX facilities via the usual R-type model specification language. The packages basically call BayesX in the background and then transfers all estimation results back to R. An alternative, manual way to access BayesX from R is the system command in R that allows to call the command line version of BayesX via
    R> system("BayesX filename")
    where filename contains a batch file with BayesX commands. Before calling system, you will have to ensure that all data sets and geographical information have been stored at the places specified in the batch file. After completion of the system call, the estimation results can be visualised with functionality provided in the BayesX R package.

  • Does BayesX work with other operating systems than Windows?
    For a long time the answer has been no, but version 2.0 of BayesX is now available in an open source distribution that can be compiled with GNU C++ compiler. As a consequence, BayesX now also works under Linux and Mac OS. A possibility to avoind manual compilation on some systems is to use R-package R2BayesX (available from CRAN) that provides pre-compiled BayesX version for example for Mac OS.

  • When running BayesX on Windows Vista or Windows 7, an UnsatisfiedLinkError appears.
    There may be two reasons for this problem:
    • The error may be caused by an internal Windows mechanism that should prevent software from executing possibly dangerous code in the background. BayesX of course does not do this but simply tries to start a new process running the actual computations in addition to the process running the graphical user interface. You can prevent this by executing BayesX as administrator, an option that you obtain by right-clicking on the BayesX icon. An alternative would of course also be to use the command line version of BayesX instead of the GUI-version. The command line version can be found in a subdirectory of the installation directory.
    • You may have to start BayesX in the compatibility mode for Windows XP. Therefore locate the BayesX application in the program folder (or an alternative installation directory you may have chosen), right-click in properties, select the compatibility folder, and tip the "Run this program in compatibility mode for:" option with "Windows XP Service Pack3". Thanks to Elisa Duarte for providing this workaround!


  • BayesX crashes. What did I do wrong?
    Of course, there is no definite answer to this question. There is the not too unlikely possibility that there is a bug in the software but in our experience, the overwhelming portion of crashes is caused by problems in the data and / or the model specification. Some possible difficulties are
    • Complete separation in binary or categorical regression models. This happens particularly frequently when considering binary covariates with only a very small number of ones or a larger number of response categories. Complete separation leads to the divergence of parameter estimates and therefore ultimately to the crash of BayesX.
    • A small fraction of excessively large counts in count data regression. Since all count data models in BayesX use the exponential link function, numerical difficulties can occur when there are some out-lying large response values.
    • Non-identifiable model specification. Again, such models lead to the divergence of parameter estimates and ultimately to a crash of the software. Examples include models where the same term has been specified more than once or varying coefficient models without appropriate centering restrictions (see the reference manual).

    In general, there are several useful strategies to identify whether the model specification is problematic for the data at hand:
    • Start with a very simple model to reduce the probability of numerical difficulties.
    • Add only one covariate at a time to identify the covariate that possibly causes the crash.
    • Stop estimation before the crash occurs, for example by specifying only a small number of MCMC iterations or a small number of maximum iterations for REML estimation. The estimation results from the model obtained may contain valuable information to identify numerical difficulties.


  • How should I cite BayesX?
    The recommended citation is "Belitz, C., Brezger, A., Klein, N., Kneib, T., Lang, S., Umlauf, N. (20xx): BayesX - Software for Bayesian inference in structured additive regression models. Version x.x. Available from http://www.bayesx.org". (with x replaced by information corresponding to the version you have used). However, please also cite the methodological publications to credit the work of the BayesX contributors.