[Lemon-user] Network Simplex with Floating Point Flows

[Péter Kovács]

Hi Samuel,

You're right, it wouldn't be too difficult to support floating-point 
input in the network simplex algorithm, but it hasn't been implemented yet.

However, you can avoid floating-point values relatively easily by 
applying an appropriate scaling of the input data. Note that 
floating-point calculations are usually risky, due to the inherent 
numeric errors. So it may be safer to scale the data and use integers, 
although some algorithms in LEMON directly support floating-point data 
using the concepts of Tolerance classes (e.g. Preflow).

If you would like to contribute, you are welcome to implement this 
improvement (in a similar way how other algorithms handle Tolerance 
classes).

Regards,
Péter



On 2017-12-04 16:47, Samuel Gerber wrote:
> Hi,
> 
> I am curious what stands in the way of allowing floating point flows 
> (capacities, supplies, costs) in the network simplex algorithm?
> Would this be difficult to add to the Lemon solver?
> As far as I understand the CPLEX implementation works with floating 
> point inputs.
> 
> Thank you
> -- 
> Samuel Gerber
> R&D Engineer
> Kitware, Inc.
> 
> 
> _______________________________________________
> Lemon-user mailing list
> Lemon-user at lemon.cs.elte.hu
> http://lemon.cs.elte.hu/mailman/listinfo/lemon-user
>