[the options menu0]    The Options Menu

Contents

    Input/Output options...
    Update options...

Input/Output options... [top]



[the options menu1]

Input options

window or file: if window is selected then input will be read from the top (focus) window; if file is selected then the user will be asked for a file to read input from..

Output options

window or log: if window is selected then a new window will be opened for each new piece of output (statistics, traces, etc.); if log is selected then all output will be pasted into a single log file.

output precision: the number of significant digits in numerical output.


Update options... [top]


[the options menu2]


Once a model has been compiled, the various updating algorithms required in order to perform the MCMC simulation may be 'tuned' somewhat via the Updater options dialog box (select Update options... from the Options menu):

methods: this "selection box" shows the system names of all updater methods currently loaded. The methods that are being used to update the current model have a star after their name . All other components of the Updater options dialog box (i.e. fields and command buttons) pertain to the currently selected item in methods .

used for: this text field simply describes what type of node the currently selected method is generally used for.

Note: it is now possible to change the sampling methods for certain classes of distribution, although this is delicate and should be done carefully - please see Changing MCMC Defaults (advanced users only) for details.

iterations: some updating algorithms entail iterative procedures that terminate when some relevant criterion is satisfied. It is always possible, however, that within a given Gibbs iteration this criterion cannot be satisfied in reasonable time. In such cases, rather than allow the computer to 'hang', it is preferable to specify a maximum number of iterations allowed before an error message is generated. This maximum number of iterations is displayed in the iterations field, which may be edited. In cases where iterative procedures are not required the iterations field will be greyed-out.

adaptive phase: some updating methods, such as Metropolis-Hastings, have an adaptive phase during which their internal parameters are tuned based on information gained from the chain(s) generated so far. All samples generated during an adaptive phase should be discarded when drawing inferences but sometimes the default adaptive phase is longer than necessary, meaning that the sampler is somewhat wasteful. (Alternatively, the default adaptive phase may not be sufficiently long to allow proper tuning.) When the adaptive phase field is not greyed-out (indicating that the currently selected method in methods requires tuning) it displays the length of the adaptive phase in iterations (Gibbs cycles) - this may be edited by the user.

over-relaxation: many updating methods are capable of generating over-relaxed samples. Here, at each iteration, a number of candidate samples is generated and one that is negatively correlated with the current value is selected - the time per iteration will be increased but within-chain correlations should be reduced. The number of candidate sam ples (including the current value) is dis played in the over-relaxation field.

Set: The Set command button applies the values shown in iterations , adaptive phase , and over-relaxation to the updating method currently selected in methods for the current model - if a new model is loaded or if WinBUGS is shut-down and re-started then the software will revert to its default values.

Save: The Save button also applies the values shown in iterations , adaptive phase , and over-relaxation to the currently selected method, but it also saves those values as defaults for that method - the next time that that method is used the new values will be selected automatically.

Disable : The disable button stops the currently selected updater method being used to update a new model. Once this button has been clicked it name toggles to Enable and clicking it enables the method for a new model.