R S D
Relational Subgroup
Discovery
through First-Order Feature Construction
- Download Page -
RSD is a system for relational subgroup
discovery in individual-centered domains, incorporating principles described in
Zelezny F., Lavrac N.: Propositionalization-Based Relational Subgroup Discovery with RSD.
Machine Learning 62(1-2):33-63, 2006 (pdf).
RSD can however be useful as well in
tasks other than subgroup discovery, as long as they require to generate
a
propositionalized representation of class-labelled relational data achieved
through constructing first-order features.
RSD has been implemented in Yap
Prolog by Filip Zelezny. No
registration is required for download, but it
would be nice of you to leave
a note on who you are, why / how you want to use RSD and what you think of
it.
NEW: Additional application sample: Mutagenesis (included in the download package).
You can download:
- The complete RSD Package: rsd.tar.gz
(465 KB)
including program codes, samples and a user's manual.
(requires Yap)
- The RSD user's manual (postscript or .pdf)
Main advantages of RSD
Syntactical feature construction controlled by mode declarations very similar to those used in the popular systems Progol and Aleph * Various user-adjustable constraints (syntactical or data-related) used by RSD's powerful pruning rules towards performance gains * Filtering of irrelevant (unneeded) features * Generation of propositionalized data representation compatible with the popular systems CN2 and Weka * Automated train-test splitting and stratified cross-validation process in RSD's rule induction component
Acknowledgement
Development of RSD's advanced propositionalization functions was supported by the project 1K04108 of the Czech Ministry of Education and by Funding of Ministry of Higher Education, Science and Technology of Slovenia grant Knowledge Technologies.