abagraph and proxdd experiments

Introduction

We here provide data for the experimental comparison of the systems proxdd and abagraph for assumption-based argumentation, as described in the paper:

• Argument Graphs and Assumption-Based Argumentation.
  Robert Craven and Francesca Toni. Artificial Intelligence, 2016.

Frameworks

The file randomaf.pl was used to generate 680 assumption-based argumentation frameworks; this program runs under SICStus prolog (versions > 4.2.x). To run this, line 39 should be edited to point to the right output directory for the frameworks to be written to. Then, load "randomaf.pl" into SICStus and use make_exp_frameworks/2; the first parameter is the number of frameworks to make per value of the various settings used to generate the frameworks, and the second parameter is an identifier which will be used in the filenames (for discriminating between different batches of frameworks).

The random generation will verify whether the frameworks generated are cyclic or acyclic, and print this information to screen; whether the frameworks contain cycles is also indicated in the filename.

For the paper experiments, we passed '10' as the first parameter. The output frameworks are contained in frameworks.tar.gz.

Results

Experiment design is covered in Section 6 of the paper. The results are collected in the file results.pl (3.2MB). Facts in the file take the form:

• result(CYCLE, EXP_TYPE, STEP, FW, QUERY, PROG, DERIV, STRAT, RES, TIME, SOLS, FILE).

where:

• CYCLE - is either "acyclic" or "cycles" depending on whether the framework contains cycles.
• EXP_TYPE - is "depvary" if the parameters vary dependently, and "indvary(#N)" where there is independent variation.
• STEP - where the framework occurs in the sequence of varying parameters.
• FW - which framework of the (possibly many) for the given parameters.
• QUERY - which sentence was chosen as the query.
• PROG - 'abagraph' or 'proxdd' depending on the proram used.
• DERIV - was the derivation admissible ('ab') or grounded ('gb').
• STRAT - which search strategy was used.
• RES - 'success' (all answers found), 'time_out', or 'failure' (an exception thrown).
• TIME - time taken, in ms.
• SOLS - the number of solutions found.
• FILE - the stem for the filename of the framework.

Author

All code written by Robert Craven, in the Department of Computing, Imperial College London.