C-Farmr homepage

Bibliography

Luc De Raedt and Jan Ramon, Condensed Representations for Inductive Logic Programming, proceedings of the Ninth International Conference on the Principles of Knowledge Representation and Reasoning, Whistler, Canada, June 2004

Download

Here is a .tar.gz file.

Documentation

Data preparation

Data should be loaded before running apriori. Also, settings and rmodes should be present.

rmodes.pl contains the rmodes
background knowledge in bg.pl

The downloadable distribution loads the mutagenesis dataset by default as an example.

Settings

Settings go into the file msettings5.pl

key/1 predicate must be defined in datafile; ?-key(K) succeeds for every example. it determines what is being counted.
:- dynamic theory/2 . in settings
minfreq(N). N is minimum no of examples to be frequent
rmode(predicate(modes...)) for each predicate; the arguments can be - (input) or + (output) or constant (specified). The rmode key is always - . The rmode of other predicates is always entirely + or entirely - with the exception of the one key argument that identifies the key type
realmode(predicate(modes ... )). for predicates that can loop
optional theory : theory(conclusion, condition) defines a background theory about the queries of interest. If no theory is defined, define theory(_,_) :- fail. to avoid sicstus errors. theory/2 influences s-free-ness
predicates([p/a, ...]) list of all predicates p iwth arity a
types([types ...]) list of types
type(pred(types ...)) for all predicates
before(X,Y). predicate X is before predicate Y where predicates of type ---- always come before predicates of ++++++++
subsume_meth(Meth). where Meth is obj_ident or theta specifies which subsumption setting is to be used: object identity or theta subsumption.
delta(off). specifies that no delta-free rules should be produced. delta(Delta). specifies that delta-free rules should be produced with Delta the value of the delta-parameter.

Output

in files f* one finds the different sets of frequent queries
in files stream.pl one finds the tested infrequent ones.
in files f*.closed one finds the closed rules.

Running

To run the system, first start sicstus and load the program:

   sicstus -l mfarmr4

Next, run apriori from the command line:

  ?- apriori(Depth).

where you should replace Depth by the actual number of levels you want to mine. After patterns have been mined, you can optionally make closed rules with

?- [closed].              % loading closed.pl
?- close_ff_upto(Depth).  % compute closed rules; Depth is the number of levels

For producing statistics the scripts run and runall can be useful.