Thoughts about generation time/process

[Chafik Graiguer]

How to interpret generation process?
the following scenario happend todady:

701 out of 717 activities placed
Elapsed time: 20 m 25 s

Max placed activities: 716 (at 9 m 01 s)

Seeing that FET was already  close to a solution, earlier at 9 munites, I canceled generation

The second attempt was good !!

717 out of 717 activities placed
Elapsed time: 9 m 12 s

Max placed activities: 717 (at 9 m 12 s)

So, can we make it as a rule:
If FET was quickely close to 100% then went down for a long time, user has to start new generation
?

[Liviu Lalescu]

Not quite :-)

I am not sure I remember exactly, but I hope I am not mistaking: there is a file, examples/Namibia/by-Bobby/set-2/may-take-hours/CONCORDIA.fet. I usually solve it by generate multiple with 3 minutes limit for each timetable. It solves in about 30% of cases in this way. I wanted to see the behavior and let it generate for a long time. It has 1502 activities. It came to 1501 in 1 minute. Then it went down. After about 1 hour (or 3, I am not sure), it solved the timetable.

I invite you to check this behavior for CONCORDIA.fet or for your file.

Also, file examples/Romania/Oradea/FS-2008-2009-difficult.fet may solve in a few minutes or in 5 hours. But it is different, because it will cycle from ~340 to 390, back and forth, until getting to all the 391 activities placed.

If you still have the random seed for your file, when it failed to solve in 20 minutes, you may want to leave it for a longer time.

We need more tests for that.

One other test would be to modify the examples/Germany/secondary-school-1/using-subactivities-constraints/german_subact_constr.fet, make max gaps per week for teachers = 2 and all min days between activities constraints have 99.75%. Sometimes, it solves fast, sometimes seems not to solve, in the same manner as you describe (I do not remember having patience to check that each time FET will solve the file).

PS: I did my best to avoid having to cancel generation and restart. I did my best to avoid cycles :-)

[Chafik Graiguer]

Interesting !!
From now on, I will generate using Random seed
To collect accurate data, I hope other users will contribute about thier own findings or tests regarding generation process

PS: I did my best to avoid having to cancel generation and restart. I did my best to avoid cycles

No doubt !!