Re: [Minion-general] switches

[Tias G. <tia...@cs...>]

On Thu, 12 Jun 2008 21:53:25 +0200, Chris Jefferson  
<ch...@bu...> wrote:

> Thank you for your comments!
>
> Fortunately, in the next couple of weeks Minion's documentation should
> improve hugely, as a Constraint Programming Summer School is being
> held in St. Andrews ( http://www-circa.mcs.st-and.ac.uk/cpss2008/ ) in
> which we will be using Minion extensively. We will use your comments
> when making these changes!

I'll be attending that summer school, that's why I'm already testing it  
out a bit on my own models : )

> As you say, the "-quiet" and "-verbose" flags only effect the parser.
> They should probably be renamed.
>
> You are also right that the "-printsol" option never does anything.
> Sorry about that.
>
> The 'PRINT' statement and any command line flags are supposed to
> forfill different roles. The 'PRINT' statement decides which variables
> should be printed, as in many problems there are lots of variables
> whose values we do not care about. This set cannot be changed from the
> command line, the only choice which can be made is how things are
> actually printed.
>
> you might want to look at the "-tableout" option, which puts a number
> of useful facts into another file, including size of search and time
> taken. Then you could seperate reading the solutions and reading info
> about search.
>
> In terms of making Minion go faster however, there are some more
> useful options available to you. You may have to experiment with
> these:
>
> ...

Thanks for the tips, but my problem uses only binary variables, and  
(reified) linear constraints on them, so bounds are not relevant and  
consistency isn't really influential.

> -varorder is another powerful tool, but can be difficult to tune. Look
> out for overtuning for one or two instances. Useful values to try are:
>
> static (default) - goes in the order you list in the 'VARORDER' section.
> conflict - which tries to search variables which have previously
> caused problems.
> sdf - Generally considered the "standard" order most people use.

Interestingly, I've compared with gecode for some datasets, and minion is  
slightly better than gecode if I use the input order as variable ordering.  
If i take 'degree_max' (take variable which is most constrained) then  
gecode becomes significantly faster.
Would your system allow for a variable ordering similar to 'degree_max' ?  
I'm interested to see if minion would outperform gecode, and how it would  
scale into 10 000's of variables and constraints.


Greetings,
Tias