Automatic algorithm configuration for
clingo
This program is free software: you can redistribute it and/or modify it under the terms of the 3-clause BSD license (please see the LICENSE file).
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
You should have received a copy of the 3-clause BSD license along with this program (see LICENSE file). If not, see https://opensource.org/licenses/BSD-3-Clause.
acclingo relies on SMAC3 to automatically optimize parameter configurations of clingo and clingo-based systems given a set of instances and a performance measure.
acclingo only requires SMAC3 to be installed, sources and documentation can be found here: https://github.com/automl/SMAC3.
No further installation is necessary and the main script can be executed via following commandline:
python scripts/acclingo --instance_dir <directory> <optional options>The directory with the instances is the only required parameter, all other options are optional, see --help for more information.
Per default, acclingo retrieves a parameter configuration optimizing runtime for solving grounded instances from the parameter space described here using instances that may be gzipped from <directory> .
We provide static binaries for Linux 64 of runsolver 3.3.4 here.
This simple example assumes that you have clingo installed. In this example we optimize the solving time for the instances inside the test-domain directory. We give it a 10 second cutoff per instance and configuration pair and a total wall time of 180 seconds. Run the following command while in the main acclingo folder:
python scripts/acclingo --instance_dir test-domain/instances --cutoff 10 --ac_budget 180 --tae_args "{\"encoding\": \"test-domain/blocks-encoding.lp\"}"When acclingo finishes running it will print out the best configuration it found. Keep in mind that this configuration might still be the base configuration given by the pcs file.
We provide the following example enabling algorithm configuration for instances with optimization statements. Instead of runtime, we optimize a cost function taking into account runtime and quality of solution relative to a best known bound.
python scripts/acclingo --instance_dir <directory> --fn_suffix ".gz" --binary <solver>
--run_obj quality --cutoff <timeout> --ac_budget <timeout>
--tae_class acclingo/tae/clasp_opt_tae.py --tae_args "{\"best_known\": \"<csv file>\"}" --fn_suffix ".gz": Only instances with extension .gz are considered in the instance directory--binary <solver>: Provide a custom binary of the solver that gets called from the target algorithm execution class (TAE)--run_obj quality: Instead of runtime optimize a cost function--cutoff <timeout>: Timeout for individual algorithm runs--ac_budget <timeout>: Timeout for acclingo--tae_class acclingo/tae/clasp_opt_tae.py: TAE that expects <solver> to be a clingo binary and grounded instances to include #minimize statements. If no solution was found or the resulting cost is bigger then the ParN score, the ParN score is returned as costs. Per default 10*cutoff. Otherwise the cost is runtime times percental difference of achieved quality to best known bound.--tae_args "{\"best_known\": \"<csv file>\"}": String in json format that passes auxiliary arguments to the TAE. The TAE expects the csv file to have instance names without extension in the first column and integers representing the best known bound in the second.The --tae_args can also be used to change the mode of clingo. By default it is set to "clasp". If the instances require an additional file(encoding) to run, you can also provide it with the --tae_args option using "encoding":
--tae_args "{\"mode\": \"clingo\", \"encoding\": \"path/to/encoding.lp\"}"