The SIO2 project
  1. The SIO2 project
  2. SIO-1797

Check for anomalies in szkopuł/oitimetool running times

    Details

    • Type: Task Task
    • Status: Closed Closed
    • Priority: Major Major
    • Resolution: Fixed
    • Affects Version/s: Current Version
    • Fix Version/s: TAG 2015/16
    • Component/s: OIOIOI
    • Labels:

      Description

      Need to check all public problems available in problemset for anomalies in szkopuł/oitimetool running times ratio.
      Especially I expect some problems to have longer running times on szkopuł than oitimetool, but most of them should have ratio around 1:1.

      Analysis of small sample (9 solutions/problems): SIO-1754

      The goal is to identify problems which timelimits should be scaled with something else than 1 and compute better ratio for them.
      Another goal is to generate chart like in SIO-1754 .
      1. oitimrunner.py
        5 kB
        Sebastian Jaszczur
      2. chart.py
        4 kB
        Sebastian Jaszczur
      1. oitimetool_times2.png
        652 kB

        Issue Links

          Activity

          Hide
          Sebastian Jaszczur added a comment - - edited
          After analysis of most of problems (around 600) indicates that maximum of 22 problems would cause problems after scaling timelimits 1:1. That's around 3%.
          It's worth noticing that 17 problems have similar problems already (but it's not the subset of previous problems; they share 5 elements).

          From those 22 problems:
          2 were some kind of bugs and oitimetool gave time of around 20 minutes when timelimit (and original runtime) was around few seconds. I don't know why this has happened, though.
          15 of them would lose max 3 points.
          3 of them would lose 5-6 points.
          1 would lose around 14 points.
          1 would lose around 23 points.

          So generally it's not so bad and with Accek we've agreed on scaling timelimits 1:1, as it's fairly OK and it's effortless. To those few problems that may lose some points - we will create a tool for searching for them, probably available from manage.py. Ticket: SIO-1819
          Show
          Sebastian Jaszczur added a comment - - edited After analysis of most of problems (around 600) indicates that maximum of 22 problems would cause problems after scaling timelimits 1:1. That's around 3%. It's worth noticing that 17 problems have similar problems already (but it's not the subset of previous problems; they share 5 elements). From those 22 problems: 2 were some kind of bugs and oitimetool gave time of around 20 minutes when timelimit (and original runtime) was around few seconds. I don't know why this has happened, though. 15 of them would lose max 3 points. 3 of them would lose 5-6 points. 1 would lose around 14 points. 1 would lose around 23 points. So generally it's not so bad and with Accek we've agreed on scaling timelimits 1:1, as it's fairly OK and it's effortless. To those few problems that may lose some points - we will create a tool for searching for them, probably available from manage.py. Ticket: SIO-1819
          Hide
          Sebastian Jaszczur added a comment -
          Raw scripts used to generate results and chart.
          Show
          Sebastian Jaszczur added a comment - Raw scripts used to generate results and chart.

            People

            • Assignee:
              Szymon Acedański
              Reporter:
              Sebastian Jaszczur
            • Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: