Re: [tablix-list] 0.2.0 configuration format

From: Nicholas Robinson <nprREMOVE@THISbottlehall.freeserve.co.uk>
Date: Fri Jan 07 2005 - 21:54:18 CET

There are some more fundamental problems:

1. More consideration needs to be given to being able to prove that the input
file is a valid description of the requirement. At the moment, we lack the
ability to confirm by any programmatic method that we have scheduled the
right number of lessons of each subject to each class and not overloaded
individual teachers. If your input file doesn't converge, you don't know if
it is because you have a mistake in the input or that the algorithm has
failed.

As this mailing list has (silently) observed, my timetable can be divided into
three bits and each bit will converge successfully with the correct number of
lessons. Running it as a whole results in missing lessons and so the
algorithm has to be slightly buggy. Relying on the output as a guide to the
correctness of the input is therefore invalid.

2. The parser needs to be improved hugely. At the moment, it accepts what it
recognises and silently throws away about 80% of what it doesn't. This leads
to a false conclusion that the input ins being executed as intended. It is
very frustrating when you find that it has been completely ignored.

3. Point 2 can be addressed by the implementation of a complete user interface
to generate the input file - but gtablix is broken for me (it has been for
about six weeks) and wasn't complete the last time I was able to use it. A
critical component that is independent and so outwith the control of the core
project doesn't seem the best design philosophy.

4. I'm still worried that Tablix can't cope with scheduling 'normal' school
timetables. Normal for the UK, at least, is a one that has every lesson
allocated for classes and around 90% of lessons for each teacher. We don't
have the luxury of 'free' or 'study' class periods to hide the limitations of
the GA. I have now completed the paper exercise to prove that my input file
is correct and matches a real timetable: yet Tablix doesn't get anywhere near
convergence.

[Aside: I've re-visited the FET project in the last few days and can get
significantly closer to convergence with this application (the equivalent of
1 mandatory error instead of 12). I'd rather stay with Tablix though because
it is intrinsically a more exciting (don't you just love the use of pvm?) and
flexible/extensible design (the modular approach), but needs must when the
Devil drives. I need something that will give me a working timetable by
Easter (2005, that is).]

5. The code needs an overhaul to improve readability and commenting - and the
documentation is sparse. I have only ever supervised work on GAs rather than
programming myself, but the difference is I was able to understand the
resulting code. Given the virtually non-existent response to mailing list
queries, having readable code and fullsome documentation are the only viable
alternatives for busy people.

Kind Regards

Nick Robinson

Fight Prejudice - Fight the Ban (see www.countrysidealliance.org)
Received on Fri Jan 07 22:15:21 2005

This archive was generated by hypermail 2.1.8 : Tue Aug 16 2005 - 20:42:50 CEST