[gecode-users] Adding restart-based search to ECLiPSe's interface to Gecode

[Kish Shen]

Hi Christian,

Thanks for your reply again!

I am reluctant to use Gecode features that are not in the official 
releases yet, as the user should be able to compile ECLiPSe and its 
open-source third-party components (like Gecode) on their own.

Currently, the released gfd interface is to Gecode 3.7. I have not 
updated the release to Gecode 4(.2) mainly because I do not want to 
remove the RESTART search engine, only to reintroduce the equivalent 
when I get RBS working with gfd.

I understand that it is easiest and perhaps best to use a common base 
type for the search engines, but in the absence of this base type at the 
moment, should I try to hide the different types of search engines 
inside my GecodeSearch class? Currently I have a singe Engine* (sengine) 
that points to the actual search engine, and a next method to obtain the 
next solution but I assume I can have different (private) pointers to 
the different search engines, and a next method that calls the 
appropriate next, depending on the search method.

I also have a related question about getting the next solution with RBS 
-- I had assumed that the next solution will be computed with a restart, 
but as there seems to be 2 search engines in rbs, with RBS controlling 
the restart on another engine doing the "actual" search (e.g. BAB), I am 
wondering if the behaviour when getting the next solution is controlled 
by this "actual" search engine, and for BAB, I don't think this is done 
with a restart (hence the need for the old RESTART engine). So is the 
search restarted when next() is called in RBS, or does this depend on 
the "actual" search engine?

Assuming that the search is restarted with next(), then can the 
functionality of the old RESTART engine be reproduced by setting the 
cutoff threshold to be very high? A related question is if it make sense 
(and if so, is there any way to do so) to use something other than the 
number of failures for deciding when to restart?

Cheers,
Kish


On 22/08/2013 08:49, Christian Schulte wrote:
> Hi Kish,
>
> My bad... There is really no common basetype for search engines, the type I
> mentioned is actually for the implementations and not the interfaces...
>
> I think you will have to wait until I have added a common basetype (there is
> a basetype, but it is not powerful enough), I might be able to do it until
> the end of the week.
>
> Sorry for the confusion!
>
> Cheers
> Christian
>
> --
> Christian Schulte, Professor of Computer Science, KTH,
> www.ict.kth.se/~cschulte/
>
>
>
> -----Original Message-----
> From: Kish Shen [mailto:kisshen at cisco.com]
> Sent: Thursday, August 22, 2013 9:45 AM
> To: cschulte at kth.se
> Cc: users at gecode.org
> Subject: Re: [gecode-users] Adding restart-based search to ECLiPSe's
> interface to Gecode
>
> Hi Christian,
>
> Thanks again!
>
> On 21/08/2013 19:17, Christian Schulte wrote:
>> Try to avoid the Gecode::Search::bab .... functions. Just create as
> follows:
>> 	Gecode::Search::Engine* e = new Gecode::BAB(...) Then it also will
>> work for RBS, and you do not have to create a second engine yourself,
>> this will the RBS class do for you.
>>
>> And do not forget to delete e eventually!
>>
>
> This is the part that I don't know how to do correct -- probably because of
> my lack of C++ knowledge, but as far as I can tell, Gecode::DFS etc are not
> of class Engine*, according to the doc:
>
> template<class T>
> class Gecode::DFS<T>
>
> for subclasses T of Space
>
> T is GecodeSpace in my case, I assume. I am not sure what class DFS is, but
> I can't see how it could be Engine*.
>
> Anway, I tried to do the assignment as you suggested:
>
>
> Search::Engine* segine;
> ...
> sengine = new DFS<GecodeSpace>(solver, o); and indeed I get the error
>
> cannot convert 'Gecode::DFS<GecodeSpace>*' to 'Gecode::Search::Engine*'
>
> (I first tried sengine = new DFS(solver, o) but got an error 'expected
> type-specifier')
>
> So I am not sure what to do -- do I need to change the type of sengine,
> and/or the way I assign an engine to sengine?
>
> Cheers,
>
> Kish
>
>
>
>