Algorithm Selection Solutions for Answer Set Programing

by M. Maratea, L. Pulina, F. Ricca

This Web site collects our tools for selecting the “best” ASP solver on the basis of some syntatic characteristics, called features, of the input program.
Algorithm selection can be applied to both phases of ASP program evaluation: Model Generation or Solving, and Instantiation or Grounding.
A complete ASP system can be obtained combining grounder+solver.

ME-ASP: MultiEngine solver for Answer Set Programing

ME-ASP applies machine learning techniques to ASP solving for selecting the best solver on a per-instance basis.

It exploits more than 50 cheap-to-compute syntactic features that allow for accurate classification of ground ASP programs written in ASPCore 1.0 (see the 3rd ASP Competition specifications).
You can select one of five multinomial classification methods to learning algorithm selection strategies.

Papers:
- Marco Maratea, Luca Pulina, Francesco Ricca: A Multi-Engine Approach to Answer Set Programming. Theory and Practice of Logic Programming, TPLP (in press) - arXiv version
- Marco Maratea, Luca Pulina, Francesco Ricca: The Multi-Engine ASP Solver me-asp. JELIA 2012: 484-487 - http://dx.doi.org/10.1007/978-3-642-33353-8_39
- Marco Maratea, Luca Pulina, Francesco Ricca: Applying Machine Learning Techniques to ASP Solving. ICLP (Technical Communications) 2012: 37-48 - http://dx.doi.org/10.4230/LIPIcs.ICLP.2012.37

Downloads:

• Optimized Version (supports only kNN):
  • ME-ASP v.0.2 (20130502)
• General Purpose Version (Java feature extractor, multiple classification methods)
  • ME-ASP v.0.1 (20120812)

Requirements:
- Linux OS
- rapidminer 4.6 (rapid-i.com)
- SUN java jre1.6.0_30 (for general purpose versions only)
- ME-ASP is a multiengine solver, please check internal engines' requirements
- NOTE: the shifter tool (see 3rd ASP Competition Tools) included in the package is needed by Gringo to process ASPCore 1.0 and requires gcj and shared library: libgcj.so.90

Installation:
- Unzip ME-ASP.zip
- Follow instructions in the README file

Release Notes:
- 2013-05-02: Introduce a re-engineered version of ME-ASP, with feature extractor written in C++, support for Gringo numeric format, and supporting only kNN classification algorithm
- 2012-08-12: Bug Fix, improved support for non-HCF programs.

Experimental ASP System combining Grounding Selector + Solver Selector

An automatic procedure for selecting an optimal combination of grounder and solver for ASP programs written in ASPCore 1.0 (see the 3rd ASP Competition specifications).

Downloads:

• Experimental Version (submitted to NMR2014):
  • ME-ASP-GR v0.1 (20140303)

Requirements:
- Linux OS
- SUN java jre1.6.0_30
- ME-ASP-GR calls external tools, please check internal tools requirements
- NOTE: the shifter tool (see 3rd ASP Competition Tools) included in the package is needed by Gringo to process ASPCore 1.0 and requires gcj and shared library: libgcj.so.90

Installation:
- Decompress ME-ASP_GR_v0.1.tar.gz
- Follow instructions in the README file

Release Notes:
- 2014-03-03: Introduce the first experimental implementation of an ASP system combining autometic selection of grounder and solver .

Grounding SELECTOR

An automatic procedure for selecting the “best” ASP grounder for ASP programs written in ASPCore 1.0 (see the 3rd ASP Competition specifications).

Downloads:

• Experimental Version (submitted to AIIA2013):
  • SELECTOR v.0.1 (20130531)

Requirements:
- Linux OS
- SUN java jre1.6.0_30
- SELECTOR calls external tools, please check internal tools requirements
- NOTE: the shifter tool (see 3rd ASP Competition Tools) included in the package is needed by Gringo to process ASPCore 1.0 and requires gcj and shared library: libgcj.so.90

Installation:
- Unzip SELECTOR.zip
- Follow instructions in the README file

Release Notes:
- 2013-05-31: Introduce the first experimental implementation of an automatic grounding selector for ASP.