commons-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Thorsten Schaefer <>
Subject Re: [math] Speeding up optimization problem?
Date Fri, 10 May 2013 18:23:49 GMT
Hi Thomas,

Thanks for the hints.

> thanks for the sample code.
> Something that may be helpful for you:
> You compute the relative action frequencies for each resource and then add
> a constraint that they have to be equal to the expected probabilities for
> this action in total. This is quite a hard constraint, try to relax it like
> that:

I tried that out and it leads to the same solution, but also the runtime doesn't change.

> Another observation that I have made, the resources array does contain a
> lot of zeros, but when creating the constraint that each resource must have
> one action assigned this is not taken into account (my understanding is
> that when in the resource array the corresponding entry is 0, this action
> can not be assigned to this resource), so I added also this:
>        if (resources[bucket][action] != 0.0) {
>          constraintValuesAction[currentIndex] = 1;
>        }
> this means only the actions that are available for this resource are used.

This is rather a matter of data representation - the assumption that the action cannot be
assigned to a resource if the value is 0 is not true, so I cannot do this optimization.

Thanks anyway for the help and I guess I just wait for the 4.0 release to be able to get the
result after x iterations. One minor remark as I think you are part of the developing team:

The type "GoalType" is currently in the package nonlinear.scalar, but also used by linear
optimizations. Maybe its a bit cleaner to move it to the optim package. 


To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message