DIPARTIMENTO DI
INFORMATICA Università di Torino | |
The following table contains a set of 451 artificial
relational problems designed in order to test the ability of a relational
learner at learning in the so called "mushy region", i.e the phase transition
in matching complexity which occurs for critical values of the number
L
of constants occurring in an example and the number
m of literals
occurring in an inductive hypothesis.
Figure 1: Distribution of the Learning problems in (m, L)-plane. Symbol (+) denotes problems solved by FOIL, i.e. for which FOIL found the correct solution or an approximation of it (Error rate < 20% on test set). Symbol (.) denotes problems where FOIL failed (Error rate > 20% on test set).
Here is a list of the files in every archive that are useful and have meaning
(all problems contain the same set of files), all other files being garbage. The signature of the predicates defined in the files is the following:
Each target concept is a unique conjunction of predicates. So, a file f0 that contains the following lines: R0(x0, x1) R1(x1, x2) R2(x2, x3), R3(x1, x3), R4(x2, x3)is to be intepreted as the following 1-clause logic program: yes(X):- brel0(X,X0,X1), brel1(X,X1,X2), brel2(X,X2,X3), brel3(X,X1,X3), brel4(X,X2,X3). Problem List: the numbers following m and l indicate the number m of literals in the target concept and the number L of constants in the examples. You can downloads the whole dataset here. |
[Information]
[People]
[Research]
[Ph.D.]
[Education]
[Library]
[Search]
[Bandi/Careers] [HelpDesk] [Administration] [Services] [Hostings] [News and events] |
Administrator: wwwadm[at]di.unito.it | Last update: Aug 23, 2011 | |