An extended transformation approach to inductive logic programming

Abstract

Inductive logic programming (ILP) is concerned with learning relational descriptions that typically have the form of logic programs. In a transformation approach, an ILP task is transformed into an equivalent learning task in a different representation formalism. Propositionalization is a particular transformation method, in which the ILP task is compiled to an attribute-value learning task. The main restriction of propositionalization methods such as LINUS is that they are unable to deal with nondeterminate local variables in the body of hypothesis clauses. In this paper we show how this limitation can be overcome, by systematic first-order feature construction using a particular individual-centered feature bias. The approach can be applied in any domain where there is a clear notion of individual. We also show how to improve upon exhaustive first-order feature construction by using a relevancy filter. The proposed approach is illustrated on the "trains" and "mutagenesis" ILP domains.

Propositionalization-based relational subgroup discovery with RSD

“…Furthermore, additional experiments with classificatory induction using the generated propositional form indicate that RSD produces features leading to high classification accuracy. Lastly, absolute-runtime comparisons with SINUS which implements the propositionalization procedure described in Flach and Lachiche (1999) and Lavrač and Flach (2001) indicate the superiority of RSD, although these figures are not conclusive due to the inherently different software and hardware used by RSD and SINUS.…”

“…The main contributions of this paper concern the transfer of this methodology to the multi-relational learning setting. The contributions include substantial improvements of the propositionalization step (compared to the propositionalization proposed by Flach and Lachiche (1999) and Lavrač and Flach (2001)) and an effective implementation of relational subgroup discovery algorithm RSD, employing language and evaluation constraints. Further contributions concern the analysis of the RSD subgroup discovery algorithm in the ROC space, and the successful application of RSD to standard ILP problems (East-West trains, King-Rook-King chess endgame and mutagenicity prediction) and two real-life problem domains (analysis of telephone calls and analysis of traffic accidents).…”

“…Utility predicates do not introduce new variables. As shown in Section 7.2, the RSD feature construction approach, described in Section 4, effectively upgrades the propositionalization through first-order feature construction proposed by Flach and Lachiche (1999) and Lavrač and Flach (2001).…”

“…LINUS performs the transformation to a propositional representation by considering only possible applications of background predicates on the (Lavrač & Džeroski, 1994) relaxed this bias so that non-constrained clauses can be constructed provided that the clauses involved are determinate. SINUS was later extended to perform propositionalization by extended first-order feature construction described in Flach and Lachiche (1999), Lavrač and Flach (2001) and Kramer et al, (2001). 10 The RSD propositionalization has the following advantages compared to SINUS.…”

Information Theory Based Feature Selection for Multi-Relational Naïve Bayesian Classifier

“…A relational database consists of a set of named tables, often referred to as relations that individually behave as the single table that is the subject of Propositional Data Mining [5]. Data structures more complex than a single record are implemented by relating pairs of tables through so-called foreign key relations [6]. Such a relation specifies how certain columns in one table can be used to look up information in corresponding columns in the other table, thus relating sets of records in the Tables 1 and 2 [7,8] .…”