<br>Thank you very much for your explanation.<br><br>I will then use scaling of values to some integer range. Any idea as to how the range will affect the computation time? That is, will the choice of a large range (in order to minimize "quantization" errors) result in much longer computation times?<br>
<br>JL<br><br><br>2013/3/13 Kovács Péter <span dir="ltr"><<a href="mailto:kpeter@inf.elte.hu" target="_blank">kpeter@inf.elte.hu</a>></span><br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi,<br>
<br>
Actually, NetworkSimplex could support non-integer magnitudes, but it would require careful modifications to avoid potential issues related to inexact computations. We did not implement this feature yet, but the reasons are only technical, not theoretical.<br>
<br>
On the other hand, CapacityScaling algorithm already supports non-integer arc costs (although it is usually slower than NetworkSimplex).<div class="im"><br>
<br>
> Would there<br>
> be any problem in "discretizing" the original problem using a large<br>
> range of integer values?<br>
<br></div>
You can do such a scaling before using the algorithm.<br>
<br>
Best regars,<br>
Peter</blockquote></div>