Re: [sumo-user] Running different scenarios with the same random seeds

[Jakob Erdmann <namdre.sumo@xxxxxxxxx>]

Hello,

I'm not sure whether I'm understanding you correctly but maybe this helps:

- you can set the random seed for sumo, duarouter, randomTrips etc. explicitly with the option --seed

- two runs of the same application with the same seed will behave in the exact same way

- If the delays between two sets of scenarios are implausible, please try to narrow down whether that is due to invalid vehicle behavior or invalid measures. You could check against the total timeLoss computed by sumo (https://sumo.dlr.de/docs/Simulation/Output.html#aggregated_traffic_measures, https://sumo.dlr.de/docs/Simulation/Output/TripInfo.html#generated_output)

regards,

Jakob

Am Fr., 18. Okt. 2019 um 17:26 Uhr schrieb Usama Shahdah <usama.elrawy@xxxxxxxxx>:

Hi Michael,

First of all thanks for your reply.

what I usually do is running sumo 10 times or more with different random seeds for duarouter to ensure random trips and different random numbers for sumo. Then estimate the average delay for each run then get the average delay for the ten runs. So I am using random seeds when running duarouter.

I do not think using only the meanTimeLossWithin alone is correct and this is based on my understanding of E3 output file here: https://sumo.dlr.de/docs/Simulation/Output/Multi-Entry-Exit_Detectors_(E3).html

as it mentioned that the meanTimeLossWithin is The average time loss collected by vehicles that have entered but not yet left the area during the written interval.

Anyway after comparing the results when using 10% LT and 30% LT of the approach demand for the 10 runs, the delay for the 30% LT is always less than the 10% LT whatever I change the cfm and lane change parameters. And this does not make sense at all. To confirm that there is something wrong I simulated the same Uturn intersection in vissim for 10 runs and the results were making sense, as the 30%LT scenarios always give higher delay than the 10% scenarios.

in vissim if I run the simulation with the same exact random seed for let's say 5 runs, I will get exactly the same results for of course the same traffic volume and same scenario.

so my question is: it is possible to use the same random seed to do the simulation, so I can compare different scenarios with exactly the same random seeds for the 10 runs. If course every run will have different random seed.

In sumo, the random seed used is a starting seed to use, so when using this same seed I got different results for the same traffic volume and same scenario. 

Hopefully I am clear this time.

Regards, 

Usama 

Hi,

I am not sure how duarouter can help here. It does not calculate any
delays. I think you should rather run many simulations with different
random seeds and then do statistics. Also why can't you use
meanTimeLossWithin directly?

Best regards,
Michael

Am 16.10.19 um 22:39 schrieb Usama Shahdah:
> Hi:
>
> I would like to compare different scenarios (for Unturn intersections)
> with different left turn movements (eg. 10%, 20% and 30% LT) using the
> same random seed for both duarouter and sumo simulation.
>
> Is there a way to do so? and how?
>
> The issue I have right now is that for high traffic volumes (1250
> veh/hr/ approach) and above, the delay for 30% LT is lower than for
> 10%LT, which does not make sense to me.
>
> Please note that I am using E3 detector to estimate the delay, and I am
> using the following formula to do so:
>
> Average Delay= (sum[meanTimeLoss * vehicleSum]?+ Sum[meanTimeLossWithin
> * vehicleSumWithin]) /( vehicleSum + vehicleSumWithin)
>
> Any help will be appreciated.
>
> Regards,
> Usama
>

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://www.eclipse.org/mailman/listinfo/sumo-user