[gecode-users] edge-finder propagator -- problem solved

[Jan Kelbel]

Hi,

finally, I found out the bug in our edge-finder propagator code -- in 
the propagation algorithm, that we previously checked more than once.
In file EdgeFinder.cpp, line 68, there should be:
	if((( tasks[i].start.min() + P + tasks[i].processingTime )) >
	   tasks[k].start.max() + tasks[k].processingTime ){ //line 21
instead of
	if((( tasks[i].start.min() + P + tasks[i].processingTime )) >
	   tasks[k].start.max() ){ //line 21

i.e. "+ tasks[k].processingTime" was missing.
For now, it computes Cmax of FT6 jobshop benchmark correctly, so we need 
to make more testing.

I forgot to mention, the edge-finder propagator I posted here only finds 
activities that should be first (so it is only half of the edge-finder).

Thanks for all your comments,
Jan

Filip Konvička napsal(a):
> Hi Jan,
> 
>> The implementation of the O(n^2) edge-finder of [Baptiste,Le Pape, 
>> Nuijten] seemed to us easier to implement than the O(n log n) algorithm 
>> of Vilim, which uses binary trees.
>> So we chose the O(n^2), as it was the first propagator we were implementing.
>>
>> Maybe we should try to implement that O(n log n) version.
> 
> The binary trees are easy - implement them using vectors: compute the 
> next larger power of two (N), make a vector of N*2-1 items, the tree 
> leaves start at index N-1, parent node of node I is (I-1)/2.
> 
> In fact, efficient computation of the next power of two is what I can 
> share with you because it is already part of the FloatVar implementation 
> I posted earlier today: it is in float/float/bi/prop_linear.icc, look 
> for "namespace __Utils".
> 
> Theta-lambda trees are usable not just for propagators, but also for 
> branchings (you can easily compute firsts/lasts using theta-lambda trees).
> 
> Hope this helps,
> Filip
> 
> 
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users