Re: [sumo-user] How to configure a simulation to obtain the same results

[Ricardo Ribeiro Gudwin <gudwin@xxxxxxxxxxxx>]

Hello all,

Thanks for all the replies so far ...

The strange thing is that using the SAME controller, and running the simulation twice, I'm getting different responses ... I am now using the following flags:
--random-depart-offset 0 --max-depart-delay 0 --step-length 0.1

My next try is to use --seed 0 and see what happens ...

My only remaining suspicion is, on the line of Michael Behrisch suggestion, that my TraCI program is a multi-threaded program ... this MIGHT explain the differences in simulation results, depending on possible threading schedules being different in each simulation run, resulting in a different decision. This is the kind of differences I am finding:

Run #1 (--tripinfo-output)

id	depart	departDelay	arrival	arrivalSpeed	duration	timeLoss	waitSteps
0	0	0	17.9	13.84	17.9	3.46	0
2	2	0	32.8	13.76	30.8	3.79	0
3	3	0	18.3	13.86	15.3	3.38	0
5	5	0	35.7	13.89	30.7	3.3	0
8	8	0	38.9	13.77	30.9	3.53	0
9	9	0	42.4	13.77	33.4	3.5	0
18	18	0	49	13.89	31	3.61	0
19	19	0	36.9	13.84	17.9	3.45	0
22	22	0	39.9	13.89	17.9	3.4	0
27	27	0	64.2	13.87	37.2	10.24	13
30	30	0	58.6	13.83	28.6	4.82	0
34	34	0	67.8	13.78	33.8	7.42	0
40	40	0	75.7	13.85	35.7	6.84	0
43	43	0	76.4	13.8	33.4	3.46	0
50	50.1	0.1	65.5	13.78	15.4	3.51	0
52	52	0	92	13.87	40	14.08	49
54	54	0	87.3	13.78	33.3	8.43	5
56	56	0	92.8	13.86	36.8	8.53	0

Run #2 (--tripinfo-output)

id	depart	departDelay	arrival	arrivalSpeed	duration	timeLoss	waitSteps
0	0	0	17.8	13.79	17.8	3.38	0
2	2	0	32.4	13.88	30.4	3.37	0
3	3	0	18.3	13.8	15.3	3.39	0
5	5	0	42.3	13.86	37.3	9.83	27
8	8	0	40.7	13.84	32.7	5.32	0
9	9	0	43.2	13.85	34.2	4.25	0
18	18	0	54.3	13.81	36.3	8.9	16
19	19	0	36.7	13.8	17.7	3.29	0
22	22	0	40	13.77	18	3.57	0
27	27	0	59.8	13.77	32.8	5.81	0
30	30	0	57.2	13.85	27.2	3.44	0
34	34	0	64.9	13.82	30.9	4.52	0
40	40	0	75.7	13.82	35.7	6.81	0
43	43	0	76.4	13.78	33.4	3.44	0
50	50	0	65.3	13.85	15.3	3.33	0
52	52	0	81.3	13.87	29.3	3.45	0
54	54	0	93.4	13.8	39.4	14.48	29
56	56	0	95.5	13.81	39.5	11.17	0

On Thu, Feb 15, 2018 at 4:42 PM, Jakob Erdmann <namdre.sumo@xxxxxxxxxxxxxx> wrote:

As soon as you introduce differences by using different traffic light controllers you potentially affect vehicle insertions as well.

This effect depends on how strongly insertions are constrained by traffic density:

- if your insertion edge is congested due a traffic jam originating from your controller then insertions cannot take place due to lack of road space

- if the traffic you insert is very dense, then slight random fluctuations in vehicle speed already affect the time, position or speed at which new vehicles are inserted (depending on insertion parameters)

A solution for fair controller comparison is to include the time by which vehicle insertion was delayed together with the time loss in the effect comparison.

That being said, in low to medium traffic situations, insertions should not be affected in a statistically relevant way.

2018-02-15 18:44 GMT+01:00 Michael Behrisch <oss@xxxxxxxxxxx>:

Hi,

Am 15.02.2018 um 18:06 schrieb Mirko Barthauer:
> Have a read in the SUMO Wiki about randomness
> <http://sumo.dlr.de/wiki/Simulation/Randomness>. You should consider
> setting the --seed parameter.

You should also look into your traci script whether it contains any
source of randomness. Sumo itself is usually deterministic, unless you
give the --random option, otherwise we would not be able to run our own
tests.

Best regards,
Michael

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

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