FAQ (updated on May. 11 PDT)

Discussion

Home Discussion Board KDD CUP 2020: Learning to Dispatch and Reposition on a Mobility-on-Demand Platform FAQ (updated on May. 11 PDT)

tqincomp

FAQ (updated on May. 11 PDT)

posted in KDD CUP 2020: Learning to Dispatch and Reposition on a Mobility-on-Demand Platform

2020-04-10 07:00

**Q: Range of coordinates in the data set?**
A: The trajectory data contains only the trajectories within the coordinates range specified in the GAIA data description.  The trip data contains trips that went through the specified coordinates range.  
The evaluation and the final testing environments cover a similar area as demonstrated in the data set.  But since they are unseen environments, there could be grids in the data set but not in the evaluation/final environments, and vise versa.

**Q: How many times can I submit through Quick Test Submission?**
A: Up to **20** times daily, including the ones that did not complete successfully due to errors in the code.  Try make best use of the Docker environment in the starting kit to clear the obvious errors in the code before submitting.  **It should be noted that the Quick Test Submission is NOT for model parameter tuning.**

**Q: Does my solution have to be an RL method?**
A: No, but you are certainly encouraged to use an RL method.  Moreover, before the Final Phase, you will be asked to submit an explanation on how your solution is related to RL in the broad sense (e.g., MDP, Approximate DP, bandits, DRL, adaptive control, optimization with long horizon, long-term value optimization, online planning).

**Q: How are the finalists determined?  How are the final scores computed?**
A: If you rank within the top 25 on either leaderboard after the final submission of the Dev Phase AND submit an explanation on how your solution is related to RL in the broad sense (also see Q&A above), you enter the Final Phase.  Your LAST submission from the Dev Phase will be run in the validation environment and the final testing environment.  The scores from the validation environment carry 40% and those from the testing environment carry 60% into the final scores, which will be used to determine the final ranks of the two tasks.

**Q: My submission failed?**
A: First, make sure that you have git cloned the latest version of the repo.  If you are submitting your own code, **make use of the software envrironment docker in the Starting Kit to make sure your code runs with the packages installed in the evaluation environment.**  Then, test run through the Quick Test Submission before you submit through Regular Submission.  We have also fixed a few minor technical issues in the last few days (prior to Apr 9).  Some of the ‘failed’ runs should have the correct scores now.  If you encounter any further problems, please let us know.

**Q: If my submission failed, would I be able to see the error messages?**
A: Yes, now you should be able to see basic error messages, but not stack traces.  If you need to see the more complete error messages, please email us (kddcup2020@didiglobal.com) the task ID.  Again, please make use of the environment docker in the starting kit to eliminate common errors before submitting.

**Q: Does the GPS trajectory data contain the part of the trips where the vehicles did not have any passenger on board?**
A: No.

23 comments

zyr17

2020-04-10 14:19

zoulx

2020-04-10 14:44

tqincomp reply zyr17

2020-04-10 18:20

Reply

0

The score on the leaderboard reflects your best score so far on that task, and it may qualify you for the final. The final score will be obtained by running your last submission from the Dev Phase in both the validation and final testing environments. So the final submission should be the one that you feel most confident about.

joy

2020-04-11 03:39

zyr17 reply tqincomp

2020-04-11 03:57

Reply

0

Thank you for your reply. So submitting 2 different answers separately for the two tasks is also impossible?

咦

2020-04-14 05:09

zoulx

2020-04-15 03:29

hhw

2020-04-18 02:19

tqincomp reply hhw

2020-04-20 20:54

Reply

0

which task_id is that?

zoulx

2020-04-22 03:07

yupei reply zoulx

2020-04-23 08:28

Reply

0

I also have this problem. Could you tell me how you solved it?

zoulx reply yupei

2020-04-23 09:37

Reply

0

os.path.join(os.path.dirname(os.path.abspath(file)), “xxx.file”) is the path for your file in the model.

zoulx

2020-04-23 10:11

FeynmanFan reply zoulx

2020-04-24 03:34

Reply

0

How did you detect you could not repose to grid “6c35a76595c7d644”? Could you please provide more info?

zoulx reply FeynmanFan

2020-04-24 03:35

Reply

0

RepositionAgentError:KeyError:6c35a76595c7d644

FeynmanFan reply zoulx

2020-04-24 03:47

Reply

0

What is the task_id please?

zoulx reply FeynmanFan

2020-04-24 06:03

Reply

0

taskId: 175780

LongXia

2020-04-26 09:50

FeynmanFan reply zoulx

2020-04-27 03:55

Reply

0

For task_id 175780, In agent.py, line 121, “values = list(self.data[str(h)][g].items()) “, the “KeyError: ‘6c35a76595c7d644’” is from here, namely there is no key=’6c35a76595c7d644’ in “self.data[str(h)].

hyporion

2020-04-27 07:42

Fengzi

2020-04-28 14:03

andrewguo

2020-04-29 05:42

zoulx

2020-05-02 07:27

zoulx reply zoulx

2020-05-02 07:29

Reply

0

<Error> <Code>NoSuchKey</Code> <Message>The specified key does not exist.</Message> <Key>176908.zip</Key> <RequestId>7A50D9103964BF5B</RequestId> <HostId> OiVYehDmoh0kJULEnUWGSBYOQ98SNo4Ep8PG45HX0rCIRapcLUT3KCwAwBzzmEtwJ1vYZ3+LQ9o= </HostId> </Error>

zoulx reply zoulx

2020-05-02 08:15

Reply

0

null Processing your submission. submission.zip May 2, 2020, 7:03 a.m. null Processing your submission. submission.zip May 2, 2020, 7:25 a.m.

zoulx reply zoulx

2020-05-02 08:37

Reply

2

Suspect that the server ran out of disk.

tqincomp reply hyporion

2020-05-06 06:03

Reply

0

The orders in the eval env have a similar geo coverage as in the released data set.

andrewguo reply andrewguo

2020-05-06 06:35

Reply

1

These four task are still waiting. One week past.

zoulx reply zoulx

2020-05-11 11:33

Reply

0

In my opinion, the global information of all idle drivers would be important for repostion task.

tqincomp reply zoulx

2020-05-11 18:11

Reply

0

As stated in the description, only a fixed group of five drivers are moved around by the reposition task. The other drivers’ idle movement is controlled by the environment.

zoulx reply tqincomp

2020-05-12 01:19

Reply

1

I think that the idle drivers controlled by the environment will influence our policy, but we do not know their status.

hhw

2020-05-26 21:10

Dingyuan

2020-05-22 12:42

The comment has been deleted by author.

zoulx

2020-05-11 11:02

The comment has been deleted by author.