Experimental Data for GHD Computation under Updates

Full raw data and other data on the experiments

The following is an overview on the contents of the data that
is provided here to allow third parties to analyze our experimental
data as well as give them all the tools needed to replicate them.

Naming of the update operations:

We initially worked with the assumption of being in a database
context, with conjunctive queries as the example application. Thus
the names of updates was different when we ran the experiments.
Instead of speaking about adding or removing constraints, the term
'relation' was used. Instead of trying to rename all our experiment
files, we simply provide a sort of translation table here:

* (in a database setting, we 'join' two relations)

Note that in the paper we consistently use the names on the right,
while here in this repository the terms on the left will be used
throughout.

Contents

• rawdata_from_condor
  
  This contains all the output from our experiments on the 44950 instances
  of our synthethic update testset. For each instance of this set, there are
  four files in the subfolder 'tosolve':  
      1)  .sh
      The script used to run it on our test machine.
      2) .error
      Where the output from stderr was saved.
      3) .log
      Where any log output from HTCondor itself was saved.
      4) .out
      The actual stdout output from our test, which was accessed
      to generate the parsed data.
   
• original_decompositions_with_mutable_subtree
  
  This contains the 44950 GHDs that were used for our update tests.
  These are _not_ GHDs of the updated hypergraphs, but instead are
  GHDs for the original hypergraph from HyperBench. They are in a
  JSON format, with indications with nodes form the minimal mutable
  subtree.  The name of the test instances as well as these
  decompositions is constructed by taking the name of the original
  hypergraph from HyperBench, and adding a suffix which indicates
  the kind of update that was performed (the naming scheme for the
  update operations was discussed earlier in this file).
   
• update_instances
  
  The 44950 test instances that were synthetically generated, as
  described in the paper. The naming scheme is the following: we start
  with the name of the original hypergraph from HyperBench (these are
  also included in repository) and adding a suffix indicating which of
  the 5 elementary modifications we performed, followed by a number to
  distinguish updates of the same kind.
   
•  HyperBench
  
  For convenience, we also provide here the hypergraphs from the
  HyperBench dataset. These graphs and more info can be found under
  its URL: http://hyperbench.dbai.tuwien.ac.at/
   
• scripts
  
  This contains the scripts needed to generate the parsed data from
  the raw data.
   
• parsed_data.csv
  
  This contains the output from the aforementioned scripts and was
  used to generate the statistics (in the form of a table and multiple
  figures) we reported on in the paper.
   
• stats.ipynb
  
  We provide as a JupyterLab notebook the queries we used to summarize
  the run time for each method. The notebook allows accessing the
  contents of the parsed data as a standard relational database and
  SQL queries, thus allowing to easily produce more statistics on the
  data, beyond what was provided in the paper.

Links to source code of used decomposition methods:

• htdLEO: https://github.com/ASchidler/htdsmt
• BalancedGoUpate: https://github.com/cem-okulmus/BalancedGoUpdate