Hi!
Can we say that at some time past FET has failed to find a solution?
An open question to everyone...
There are two files in the official FET examples which can solve in 3 minutes or in 5 hours (on my computer). The files are examples/Namibia/by-Bobby/set-2/may-take-hours/CONCORDIA.fet and examples/Romania/Oradea/FS-2008-2009-difficult.fet.
Also, some users reported more than 24 hours for a file.
May be the following constraints that are at issue:
1/ same starting time with 100%
2/ min days between activities with 100%
I think the first one(same starting time) needs better optimization, maybe, I do not know ...what do you think?
I am not sure I understand, but I can tell you that they are optimized the best I could.
mmm... May be the weight 100%?
You can use weight less than 100%, but the constraint might then be broken.
In this case the conflicts will be near and the production time will be reduced. But the quality will be reduced too!
Can the production time be reduced with a better quality of a timetable?!
Quote from: Benahmed Abdelkrim on February 26, 2017, 06:04:34 PM
In this case the conflicts will be near and the production time will be reduced. But the quality will be reduced too!
Can the production time be reduced with a better quality of a timetable?!
I guess the answer is no.
You mean that we can not avoid conflicts, if we want a reduced production time?
It is logical that the generation takes longer if the timetable is more constrained. But in some cases it can take less time. For instance, for a locked timetable, it takes very little.
Sometimes With the same schedule and under the same conditions of data and constraints, different production times are obtained...
Yes, of course, the process is semi-randomized.
...Here an example where FET takes 3H 37Min 55S to produce a timetable
Quote from: Liviu Lalescu on February 26, 2017, 11:08:06 AM
There are two files in the official FET examples which can solve in 3 minutes or in 5 hours (on my computer). The files are examples/Namibia/by-Bobby/set-2/may-take-hours/CONCORDIA.fet and examples/Romania/Oradea/FS-2008-2009-difficult.fet.
Also, some users reported more than 24 hours for a file.
Sory! But you meant that time depends on hardware; Ie that the same file can take diferends times on diferends computers?!!!...
Yes.
You can start with the same random seed to obtain the same timetable.
Quote from: Liviu Lalescu on February 28, 2017, 07:29:30 AM
You can start with the same random seed to obtain the same timetable.
Is the random seed having an effect on the production time?
Yes. Sometimes more important, sometimes less important. Please read the Help button on the Random seed dialog.
For a very very difficult timetable, and if fet fail to finish the schedule after a long time-(more than 24 hours)-, can it know that the timetable is impossible, giving an alert to the user, or stopping production...
This is not possible. The user must appreciate by himself this fact.
more than 24 hours doesn't prove that a timetable is impossible.
i also have got a very difficult timetable (with a lot of "tweaks" to speedup).
result: it even doesn't solve after a week of generating.
all the past years i always just waited and got a result. in worst case after a few hours. but this year the timetable was much more difficult. (at my old school the dataset was much easier. i always got better results after a few minutes; but my new school has very difficult dataset :-( )
but this time i changed my way how to generate.
i generated multiple timetables, but set timelimit to max 20 minutes.
result:
in most situations i got more than 90% of placed activities in less than 10 minutes. (because i highly tweak it. without that teaks i need (most times) hours to get close to 90%).
(but like i said: 1 run even didn't solve after a week! i also stopped a fistful generations after several hours without an result)
i generated a few hundreds of timetables by that and i got a few solutions in just 5 minutes.