the weight of the weight?

[Benahmed Abdelkrim]

Here are 2 examples of the same database, the difference is the weight of min days between activities.

the first;
      weight = 100%;
      production time = 2H 6min 54s,
      number of conflicts = 0
     random seed: X=105528721, Y=662445328

the second;
      weight = 95%,
      production time= 5min 33s.
     number of conflicts = 9;
    random seed: X=2105598204, Y=300887750


in the 2 examples FET has produced very good timetables. but the difference is the time!

to you to make the conclusion?

(the version used is the Moroccan version)

[Volker Dirr]

yes. that is ok.

[Liviu Lalescu]

I will run them as well.

Would you like me to add them to the next Morocco version, as example files?

[Benahmed Abdelkrim]

I will run them as well.

Would you like me to add them to the next Morocco version, as example files?

I agree

[Liviu Lalescu]

For the file which has weights under 100% you can try starting with X=123, Y=123. It took 30 minutes 12 seconds on my computer, which seems similar to your computer (because starting with X=2105598204, Y=300887750 it took ~5 minutes).

It is sometimes luck.

[Benahmed Abdelkrim]

For the file which has weights under 100% you can try starting with X=123, Y=123. It took 30 minutes 12 seconds on my computer, which seems similar to your computer (because starting with X=2105598204, Y=300887750 it took ~5 minutes).

It is sometimes luck.

your computer is faster than mine, it took 32 min.

yes the time is relative, and it appears that we can not control it. it depends on luck!!!

[Benahmed Abdelkrim]

I did another test with the first file weight = 100%. the result is surprising, the time is 43min.
I used the same random seed of the second file cited above.
Conclusion: the random seed has an impact on the speed of generation.

[rodolforg]

That's for sure, Benahmed.
As far as I understand:
1. the algorithm randomly places an activity in the timetable,
2. then checks if there is any constraint conflict;
3. if a constraint is broken, it randomly chooses to remove one of conflicting activities

[Liviu Lalescu]

Corrections:

1. It places recursively an activity so that the conflicts with the already added activities are minimum (but at level 0 of placing it might break this rule, to avoid cycling).

3. It removes ALL conflicting activities.

[rodolforg]

Thank you for clarifications, Liviu.

[Benahmed Abdelkrim]

surprise or good news, the first file with 100% weight is resolved in record time; 12 min 43s.

same random seeds of the second file cited above :
X=2105598204,
Y=300887750

only one small change: the activity ID = 533 and ID = 534 fixed in time by the constraint: acttivity has a preferred starting time instead of the constraint: activity has a set of preferred starting time.

I attach below the file and a screenshot of FET showing the end of generation

[Liviu Lalescu]

1) I examined the previous file and this file. The activities 533 and 534 have the same constraint for them, activity preferred starting time.

2) It is only luck. You can start the generation with any random seed, obtaining any generation time. If the file is only slightly different, you will obtain very different running behaviors even with the same random seed.

I am sorry that I cannot explain better.

[Benahmed Abdelkrim]

1) I examined the previous file and this file. The activities 533 and 534 have the same constraint for them, activity preferred starting time.

in the previous file the constraint used is: activity has a set of preferred starting time.
in this last file the constraint used is: acttivity has a preferred starting time.

I think that the two constraints are different.
the first constraint provides several time intervals for 533 and 534, however, the second provides only one interval for each activity.

[Liviu Lalescu]

Oh, indeed, sorry!

So, you are saying that even if the file is more constrained, it is solved faster. There are two reasons:

1) It is easier. Think of a solution file (totally locked file). It solves instantly. It might be helpful to FET to only place this activity in a single slot.

2) It might be just luck.

[Benahmed Abdelkrim]

I think he has a third possibility, maybe it's fiction but it exists (I think); to find the right combination of random seeds for each file? an equation of chance ?!