Experimental Data for log-k-decomp

Raw Data for Experiments, parsed data from Experiments, and SQLite database with the data
------------------------------------

This repository contains updated data based on a new version of log-k-decomp, and additional experiments, as they are presented in the TODS paper "Fast Parallel Hypertree Decompositions in Logarithmic Recursion Depth".

We provide the following items:

•  the complete raw output of all our experiments, directly from our test machine and run via HTCondor. Note that we have two versions here: one is raw data from our experiments for the PODS version of this work, and the other has a set of expanded tests that were done while working on the extended version that was presented in TODS.
   (rawdata_from_*.zip)
• we also provide the instances from HyberBench that we ran our experiments against.
  (hyperbench.zip)
•  to make this more useful, we also provide the extracted information from each run of each decomposition method. (parseddata_csv.zip and/or parseddata_sql.zip)
  The scripts used to parse the output is also included. (scripts_to_parse_raw_data.zip)
  
  Each test run is broken down into the graph from HyperBench its run on, the width parameter that was used (where applicable), the time the instance took to run, whether a timeout was encountered (1 hour for all runs, except for a specific test run with htdLEO as indicated in the paper where a timeout of 10 hours was used). In addition to this, the column "Correct" indicates whether the test produced an HD ("True") or whether the decomposition method determined that the graph has an hypertree width higher than the sought width and couldn't find an HD ("False"). We also provide the exact command line arguments for each test in the "RunInfo" column. All this information is contained in the "Run" table.
•  To allow easy analysis of the data, an SQLite3 database is included, already including all the parsed data. (logkExperiments.sqlite)
  Furthermore, we also include annotated queries, used to produced any of the statistics shown in the paper, such as the data for the main table, and any of the figures. (queries.zip)
• We also performed additional experiments to test the scaling of log-k-decomp when using varying number of CPU cores. The parsed data from these experiments can be accessed in form of an SQLite database in the file 'ScalingTests.zip'.

We hope this allows any interested party to easily reproduce our experiments.

Links to source code of used decomposition methods:

• htdLEO: https://github.com/ASchidler/htdsmt
• NewDetKDecomp: https://github.com/TUfischl/newdetkdecomp
• log-k-decomp: https://github.com/cem-okulmus/log-k-decomp

Errata

* in the file parseddata_csv.zip/Run.csv, there are entries for the hypergraph "rand_q0135.hg" that indicate it has hypertree-width 2. This is not correct, and is due to a manual copy error when merging test runs from previous papers (to allow comparison with results from older algorithms). "rand_q0135.hg" has hypertree-width 3, and any Runs that contradict this should be ignored. We will upload a new version at some future point in time,  once we carefully checked the entire dataset for any other such errors.