Directories

Directories at the top-level are divided between generic and model specific. The former have code that implement the MCMC algorithm and assist in making the running of experiments an easier task, while the latter hold programs that are only relevant to individual or classes of model spaces. In adddition directory pbl contains scripts for re-running experiments that have appeared in various epublications.

Directories in pbl/:

uai01/ Experiments from [Angelopoulos CussensAngelopoulos Cussens2001]

ijcai05/ Experiments from [Angelopoulos CussensAngelopoulos Cussens2005]

The generic directories are:

auxil/ Auxiliary programs such as creating counts and visits for the chain. Also for sampling random models over an SLP. See Section 8.1.

src/ The Prolog source files that implement the MCMC algorithm. Includes a lib/ directory of supporting predicates.

The model specific directories are:

bns/ Files only pertinent to MCMC over BNs. It is worth noting that most SLPs, in directory slps, are failure-free.

bns/rpdags Programs for running MCMCMS over Restricted Acyclic Partially Directed Graphs. This is a class approximating Markov equivalence classes of BNs. It shares many programs with BNs, thus it is a subdirectory of bns/.

carts/ Programs for running experiments over C&RT models. Two radically different priors and two data sets have been explored.

tutorial/ Files for the tutorial described in Section 2.

In each of the model specific directories, there might exist subdirectories similar to the top-level generic ones: src/ and auxil/. In addition the following directories have been used:

data/ This is where programs try to find the (training) data.

runs/ Repository of scripts. Example files defining run_data/11, which is used by run/1.

slps/ Where SLPs are looked for. Generally speaking, we keep all the files for constructing models here. These will normally also include the Prolog programs on which the SLPs are based.

See files Readme.txt within each model-specific directory for details on running experiments with the respective models. Also within the directories (bns/, bns/rpdags/, carts/bcw/ and carts/kyphosis/ e.t.c.) the user can find examples of Prolog code that uses run/1 (see files run_test.pl and run_disp.pl). These files ensure appropriate files for experiment-runs, likelihood, backtracking proposal, displaying and post-processing, are loaded before run/1 is called. The run_disp.pl version allows visualisation of current and proposed models. run_test and run_disp put their output files into out_test and out_disp respectively. In few cases we have created run_graphs scripts. These are normally identical to out_test but for creating a pair of plots for each experiment run. These are the stays and visits plots of the chain, which are simple indicators of how much the chain moved in the space of models.