Questions regarding the Jams task

Questions, answers, discussions related to IEEE ICDM Contest

Questions regarding the Jams task

Postby jingruih » Fri Jul 23, 2010 4:22 pm

I have some clarification questions regarding the Jams task.

1. Since a segment is said to be jammed if average speed over the last 6 minutes was lower than 5 km/h and the number of cars that have passed or stay on this segment is larger than 10, does it mean that no jams could happen in the first 6 minutes of each simulation?
2. If road segment A is ONLY connected with road segment B, and road segment B is having road work, what will the vehicles passing A do? Will they stay at the end of A forever, or will they come back from the other direction of A?
3. In the introduction part, it is mentioned that the data cover only major roads that constitute at most 10% of the whole road network. However, based on the average maximum velocity given in street_graph.txt, there are 8631 major roads, which constitute 24.5% of the whole road network. Am I missing anything here?

Thanks a lot!
jingruih
 
Posts: 2
Joined: Wed Jul 07, 2010 5:28 pm

Re: Questions regarding the Jams task

Postby banbar » Mon Jul 26, 2010 6:37 pm

Hi,

With agreeing on the previously asked questions, I would like to extend the questions with the followings:

1. How we can interpret the links whose starting and ending nodes are same? (e.g. 249476123_249476123, 251765120_251765120) If they define a junction, then what is the difference between those and the links having a length of 3 meters -which also indicate a junction?
2. a. Does each link have a finite capacity (i.e. max flow)?
b. If so, can the exit flow exceed this capacity? (does your model allow, segments having more cars than they could have)?
3. Is the distribution of start and destinations points, and the flows between them identical for all simulations?
4. What travel time model is used?
5. How, exactly, do you calculate "average speed"? Which vehicles contribute to the calculation? -The ones that entered and exit the segment within the 6 minutes or the ones that only entered the segment in that 6 minute period?

Thanks,
banbar
 
Posts: 4
Joined: Mon Jun 14, 2010 12:21 pm

Re: Questions regarding the Jams task

Postby pawelg » Tue Jul 27, 2010 12:24 pm

Hi!

Ad. 1: Yes, no jams could happen in the first 6 minutes.

Ad. 2: Drivers know about road works before start and choose routes which always avoid road works. So segments connected ONLY to segments with road works (or which lead always to segments with road works) will be always empty.

Ad. 3: The value 24.5% (8631 segemnts) is correct.

Best regards,
Pawel Gora
pawelg
 
Posts: 13
Joined: Fri Jun 25, 2010 9:12 pm

Re: Questions regarding the Jams task

Postby pawelg » Tue Jul 27, 2010 3:35 pm

Hi!

@banbar:

Thanks for the questions. We can't answer them fully, because some issues are a part of the competition task and the answers should be found by participants.

1. We confirm that there are 5 nodes which are connected with itself. It is a bug in the OpenStreetMap data.
2a. Sure, otherwise there would be no jams ;)
2b. No
3. See the task description
4. It's close to the real-world model of traffic.
5. The average speed is a harmonic average of velocities of all vehicles that entered the segment within that 6 minute period.

Best regards,
Pawel Gora
pawelg
 
Posts: 13
Joined: Fri Jun 25, 2010 9:12 pm

Re: Questions regarding the Jams task

Postby banbar » Tue Jul 27, 2010 10:32 pm

Thank you Pawel,

I have one more question related with quality of prediction results. I used the given formula to calculate the accuracy values for training data (predicted and observed values are same -perfect prediction), I realized the average quality was 0.1585 with highest quality of 0.4082 (for 1651., 3227,. 4906. samples).

If these are the perfect results, then how could we see quality values which are higher than these at the leaderboard?
banbar
 
Posts: 4
Joined: Mon Jun 14, 2010 12:21 pm

Re: Questions regarding the Jams task

Postby Swietlicka » Wed Jul 28, 2010 4:46 pm

Hi,

The quality measure is a kind of average precision, so the maximum value (for perfect prediction) is 100.

Note that the formula in task description only relates to a single sample - the overall score is the arithmetic mean of those values.

Joanna Świetlicka, TunedIT
Swietlicka
 


Return to IEEE ICDM Contest: Road Traffic Prediction for Intelligent GPS Navigation

Who is online

Users browsing this forum: No registered users and 1 guest

cron