Answers inline... Am Fr., 15. Mai 2020 um 11:00 Uhr schrieb Tetris < schmel...@campus.tu-berlin.de>:
> Hello, > > the GEH of routeSampler.py compares the turncounts.xml with the computed > rou.xml, doesn't it? > Does this GEH <5 for more than 85 % already mean that the flows are > consistent? Or are there more GEH measurements necessary for a whole > simulation? routeSampler computes a GEH for every edgeRelation for one > hour? > The GEH is reported for every data interval separately. Depending on your input these don't have to be hourly values. If there is more than on interval, you will also get aggregated statistics on all intervals. > My input data were traffic counts that were made at different times (same > time of day but different days, month and even years) so theses counts > usually shouldn't fit and need to be scaled. Does the GEH give me feedback > whether or not the turncounts do now fit and are consistent? > The only feedback you get is in the form of overflow/underflow. This can be due to inconsistent data or bad input routes. However, you can also fail in another way: If you have lots of short routes in your route input (i.e. a routes that pass only one intersection) then the sampler will always find a perfect solution even though the input data is inconsistent. > Also I don't understand the message for underflow and overflow. What does > min and max in this context mean? And Q1, Q3? > min and max are minimum and maximum underflow of all data locations. Q1 and Q3 are quantiles (https://en.wikipedia.org/wiki/Quantile) > > > Besides, if I leave out the optimization option I get a GEH <5 for 100%. > Both are missing the route with count 8. Without the optimization there > are > more vehicles in total in the simulation. So the simulation is "better" > without optimization? It is closer to my turncount.xml which has a total > count of 13251 vehicles, so only the 8 vehicles of the missing route are > also missing. > The optimizer also tries to reduce the number of routes (thereby preferring longer routes). Right now the optimizer is solving an ILP-Problem via relaxiation ( https://en.wikipedia.org/wiki/Integer_programming). This means it may return decimal numbers where only integer results are wanted. When these numbers are rounded off this reduces the GEH (even though the solver found a perfect solution for the relaxed problem). Due to the reduced number of routes, the solution is usually better in quality. (A solution would to upgrade the algorithm so it finds an exact solution to the ILP and is never worse than plain sampling). regards, Jakob > > > > > -- > Sent from: http://sumo-user-mailing-list.90755.n8.nabble.com/ > _______________________________________________ > sumo-user mailing list > sumo-user@eclipse.org > To unsubscribe from this list, visit > https://www.eclipse.org/mailman/listinfo/sumo-user >
_______________________________________________ sumo-user mailing list sumo-user@eclipse.org To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user
