Just some small things that were handy during the experiment phase:
logging for the DROP heuristic. It sometimes takes ages, being able to track its progress helps impatient people (me)
a pickle option for the benchmarked workload. This makes the examination of the results via python much easier
sorting the output for the workload's column methods. I introduced that for better reproducibility but I don't know whether it really makes a difference and if we want it
Just some small things that were handy during the experiment phase: