Data of the PhD thesis "Merge-and-Shrink Abstractions for Classical Planning: Theory, Strategies, and Implementation"

This data set contains raw data and parsed data of all experiments [1] run for the PhD thesis. They were generated using lab (see https://doi.org/10.5281/zenodo.399255).

The raw data files (sievers-phd2017-raw-data-part*.tar.gz) contain a subdirectory for each experiment, each containing a subdirectory for each planner run of the experiment, distributed over the directories runs-*. For each run, there are the input PDDL files, domain.pddl and problem.pddl, the compressed output as generated by the translator component of Fast Downward (output.sas.xz), the run log file "run.log" (stdout), possibly also a run error file "run.err" (stderr), and the run script "run" used to start the experiment. The latter cannot be directly used, however, because the directory containing source code and build (compiled object files) have been removed for space reasons. The code is publicly available under https://doi.org/10.5281/zenodo.1163381. The (lab) scripts for parsing run.log are also available in the main directory of each experiment. All other scripts and a corresponding lab version are available on request.

For each raw data experiment, the parsed data file (sievers-phd2017-parsed-data.tar.gz) also contains a directory of the same name, with "-eval" appended. It contains a single file called "properties" that combines all of the experiment's parsed data (which can be and was generated from the raw data using lab and the parser scripts). They are in the json format and can be used for easy manipulation of the data. The directories with the prefix "paper-" and "talk-" are combinations of other directories (using the "fetch" mechanism of lab). It is recommended to use these, because due to technical errors, the original eval directories do not contain all runs of all planners (to be more precise: they contain all runs, but a subset of the planner have not been started in these experiments for technical errors and thus considered not solving the task). The missing ones have been run separately, see the directories with "missing-runs" in their name. This is also the reason some of these directories ("paper-", "talk-") contain files named "old-properties" and "fixed-properties" besides the actual "properties". "old-properties" are those with missing/faulty runs, "fixed-properties" are as "old-properties", however with the data of faulty runs removed, and "properties" are as "fixed-properties", however with the addition of the fixed missing runs (in fact, these always contain *all* fixed missing runs of all experiments, for technical reasons).

The file sievers-phd2017-parsed-data-all-and-random-merge-strategies.tar.gz contains parsed data of earlier experiments (see [1]), for which no raw data has been archived. The directories contain properties files in the json format.

[1] except raw data for the parsed data "sota-symba-spmas-eval" (which in the meantime was added to a separate data set available under https://doi.org/10.5281/zenodo.1189912) and all re-used experiments from the paper "An Analysis of Merge Strategies for Merge-and-Shrink Heuristics" (Silvan Sievers, Martin Wehrle and Malte Helmert, ICAPS 2016), for which the raw data was too large to be archived.