[gecode-users] Search statistics
Christian Schulte
cschulte at kth.se
Mon Jun 9 11:25:11 CEST 2008
As said, the most important is failures: that describes how much search you
use and is a good measure how difficult your problem is (or if you compare
different models of the same problem, how good a model is: less search,
better model, provided runtime is better and memory does not explode).
Then, of course: runtime and memory.
The rest does not matter.
Christian
--
Christian Schulte, www.ict.kth.se/~cschulte/
-----Original Message-----
From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
Of Malcolm Ryan
Sent: Monday, June 09, 2008 11:18 AM
To: gecode list
Subject: Re: [gecode-users] Search statistics
I'm writing a paper about a constraint problem I'm working, comparing
a couple of different representations. What statistics are appropriate
to compare?
Malcolm
On 08/06/2008, at 8:58 PM, Christian Schulte wrote:
> Hi,
>
> here we go:
>
> - propagates: how many propagators implementing constraints have been
> executed.
>
> - clones and commits: search uses recomputation and cloning. Only
> every so
> often a clone is created (increasing the number
> of clones reported). When failure occurs, a new node (space) is
> recomputed: it is searched for the lowest copy in the
> search tree. Then an additional copy is created from that copy and
> recomputation is used to recompute the node that
> is actually node as next search node. For recomputation, the
> decisions
> made by branching are replayed: each decision
> replayed amounts to a commit operation.
>
> With other words, commits give how much effort has been spent on
> recomputation and clones how much effort has been spent on
> recovering from
> failure (more or less).
>
> But the real meaningful numbers are the number of propagations and the
> number of failures, the rest is more implementation specific.
>
> I hope that helps
> Christian
>
> --
> Christian Schulte, www.ict.kth.se/~cschulte/
>
>
> -----Original Message-----
> From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On
> Behalf
> Of Malcolm Ryan
> Sent: Thursday, June 05, 2008 5:40 AM
> To: gecode list
> Subject: [gecode-users] Search statistics
>
> What are the precise meaning of the Clones, Commits and Propagates
> values given by a gecode.Statistics object?
>
> Malcolm
>
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users
_______________________________________________
Gecode users mailing list
users at gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users
More information about the gecode-users
mailing list