Random seeds

Benahmed Abdelkrim · March 29, 2018, 05:21:45 PM

Hello!

Random seed appears to affect production speed. Is it a way to give it proper values to get a better output SPEEDwith the same dataset?

Liviu Lalescu · March 29, 2018, 05:24:07 PM

Absolutely no way

Volker Dirr · March 29, 2018, 05:27:10 PM

The seed is random, because we don't know a good value for the first start. There is no way to "calculate" a good seed (i think). It is "chaotic". A good seed for one timetable might be a very bad seed for an other timetable.

Benahmed Abdelkrim · March 29, 2018, 05:46:10 PM

I think from some tests it is better to have one odd number and the other an even number!
Other tests are required.

Benahmed Abdelkrim · March 30, 2018, 10:10:48 AM

I can confirm that for the same dataset and for different values of random seeds we get different production time.
He got it with me. The program produced the table within 6 minutes and then, when it was released again the production time exceeded 60 minutes!
Is not that a bit strange!?

Liviu Lalescu · March 30, 2018, 10:33:53 AM

We have also Namibia/by-Bobby/set-2/may-take-hours/CONCORDIA.fet and Romania/Oradea/FS-2008-2009-difficult.fet which can take 3 minutes or 5 hours.

It depends on luck.

Benahmed Abdelkrim · March 30, 2018, 02:48:40 PM

So time is just a matter of luck and has nothing to do with other things !?
Dataset, Constraints, Weight ...

Liviu Lalescu · March 30, 2018, 02:53:59 PM

It has. For the Germany/secondary-school-1/ it solves in comparable times.

At some stage, I order randomly the times and choose the first one which is part from the best, meaning conflicting less.

Benahmed Abdelkrim · March 31, 2018, 08:54:55 AM

I think that the method of introducing the students' structure and some activities of each dataset has an impact on the speed of production and, of course, some time and space constraints.
Some activities are arranged by the program in the last order but take considerable time to schedule them!. And in this context I assure you that it lasted 5 hours and 15 minutes to complete the last activity !?

Benahmed Abdelkrim · March 31, 2018, 09:44:44 AM

I attach below an example(with morocco33) compressed with random seeds before.

FET takes 6 hours 30 minutes for completion.

the last activity took alone 5 hours 15 minutes!?.

Liviu Lalescu · March 31, 2018, 09:50:33 AM

I cannot prove that the algorithm will finish. I can only hope. And sometimes it takes very long.

Benahmed Abdelkrim · March 31, 2018, 10:17:31 AM

When he goes back and then forward, is not this an indication that he will complete the table, and that it is only a matter of time

hmm... FET, did not deliver all its secrets.

Liviu Lalescu · March 31, 2018, 10:19:02 AM

Quote from: Benahmed Abdelkrim on March 31, 2018, 10:17:31 AM
hmm... FET, did not deliver all its secrets.