Arguments for run_data/11

1:ID unique id for the experiment. Usually a small positive integer.

2:BP backtracking proposal to be used. One of

bp(P) probabilistic backtracking, with fixed probability; as presented in [Angelopoulos CussensAngelopoulos Cussens2001].

cp(D) cyclic probabilistic backtracking, of focus $D$; as presented in [Angelopoulos CussensAngelopoulos Cussens2001].

cd(S) cyclic deterministic backtracking; cycling $k$ from $1$ to $S$. Where $k=1$ is the root of the SLD-tree.

uc uniformly chosen backtracking, which chooses amongst the available points.

3:SLP the location of the source file for the SLP to be used. Extension .slp is not necessary and files will be looked for at current and ./slps/ directories.

4:MSD model space identifier. Can either be set to `rm' or to `fm' (see Appendix C). The former identifies real model spaces- upon failure to reach a model usual Prolog backtracking is employed to explore alternative choices, in a exhaustive shallow-first search. The latter allows failure models to be returned at top level of the MCMC. When this is the case, there will be no attempt to jump to the proposed, imaginary, model. Instead, the count of staying with current model will increase by one.

5:DataCall a callable term. This argument performs two roles. Firstly, file $DataName${.pl}, the predicate name of DataCall, is compiled. Secondly, DataCall is called as a Prolog goal before the main MCMC sampling begins. It is therefore, a `hook' which can be useful for any needed pre-processing. DataCall is normally defined in $DataName${.pl} and it should at least succeed. It may have shared arguments with $6:Query$; so data may provide parameters of the space of all possible models (e.g. C&RT in directory cart/). Optionally, `clean_up_'$DataCall$ may also be defined in $DataName${.pl}. This is only necessary when a single run experiment uses more than one data set. See file bns/data/jc_asia_data_dyn_all.pl for an example.

6:Query a $Name/Args$ structure, where $Name$ is a predicate symbol and $Args$ is a list. If $L$ is the length of the list $Args$ then Name/(L+1) is the top-level predicate for the chain and the predicate should be defined in $SLP$. The last argument of calls to Name/(L+1) will be a fresh variable. The software expects the this variable to be instantiated to a model each time the call to Name/(L+1) succeeds. The first $L$ arguments in calls to Name/(L+1) will be those supplied in the list $Args$.

7:Contr an atom that will be added to the stem filename of the output files created from this experiment.

8:Repeats a large integer, stating the length of the Markov chain.

9:HotChnsIds a list of hot chain identifiers. For power tampering, which is the only kind currently supported, each identifier should be a power $\beta_j$ with $0 < \beta_{j} < 1$. See Section 8.3.

10:RndSeedIdx a random seeds index: an integer between 1 and 1000, see file auxil/seeds.pl.

11:Exts the extensions to be created from the main output file. This can be used to restrict the files generated from applications of exec_extension/2. The safer value is `all'. Alternatively a list of extensions matching the first argument of defined exec_extension/2 clauses can be given. The empty list, and equivalently `none' are also acceptable values which create no extention files.