{ "id": 5891, "title": "Few feedbacks just after the milestone..", "url": "/competitions/flight2-milestone/discussion/5891", "author_name": "", "post_date": "2013-09-26T11:01:46.943Z", "votes": 2, "comment_count": 9, "views": 4150, "content": "

I must say - so far, the competition has been really fun!! I'm looking forward to the phase 2 and the milestone winner’s ideas! :)

\n

Below a couple of questions...

\n

actualTakeoff & actualLanding: these files (should) contain the number of planes arriving and leaving the specific airport at each time. They can be easily recreated on the training data looking at the actual runway arrival, actual runway takeoff from the flight history files.
In the test set, we’re not given any scheduled plane leaving after the cutoff time, which is kind of legitimate to have, as plane are schedule days ahead. So we’ll have no idea of what else should be in the air afterwards. Can we have a list of the scheduled flights after the cutoff, without the actual landing & takeoff? Or, is already provided and I'm just blind?

groundCondition: these files contain the airport conditions at each point of time after the cutoff. In the oneday simulator file, some airports don’t have weather informations for the 1st hour or after the 4th hour, which will result in giving the default values (10 visibility, 45 temperature) for any planes landing before the 1st or 4th hour at these airports. This turns out to be a disadvantage if we can forecast accurately the weather for every hour (or less) for each airport. Can we have some insights on how these files are created?

", "messages": [ { "id": "31591", "postDate": "09/26/2013 11:01:46", "content": "

I must say - so far, the competition has been really fun!! I'm looking forward to the phase 2 and the milestone winner’s ideas! :)

\n

Below a couple of questions...

\n

actualTakeoff & actualLanding: these files (should) contain the number of planes arriving and leaving the specific airport at each time. They can be easily recreated on the training data looking at the actual runway arrival, actual runway takeoff from the flight history files.
In the test set, we’re not given any scheduled plane leaving after the cutoff time, which is kind of legitimate to have, as plane are schedule days ahead. So we’ll have no idea of what else should be in the air afterwards. Can we have a list of the scheduled flights after the cutoff, without the actual landing & takeoff? Or, is already provided and I'm just blind?

groundCondition: these files contain the airport conditions at each point of time after the cutoff. In the oneday simulator file, some airports don’t have weather informations for the 1st hour or after the 4th hour, which will result in giving the default values (10 visibility, 45 temperature) for any planes landing before the 1st or 4th hour at these airports. This turns out to be a disadvantage if we can forecast accurately the weather for every hour (or less) for each airport. Can we have some insights on how these files are created?

", "rawMarkdown": "", "votes": null }, { "id": "31602", "postDate": "09/26/2013 12:43:47", "content": "

I second the request; having some useful information on how these files are created - at least assuming a full dataset - would provide a more complete environment to work with.

\n

Regarding scheduled flights after the cutoff, I may be wrong, but I think we have this information in flighthistory.csv. What we're missing is the actual takeoff/landing time if that happened after the cutoff. What I'm wonder is if the flights we have to compute routes for are included among those in actualLandings.csv

", "rawMarkdown": "", "votes": null }, { "id": "31607", "postDate": "09/26/2013 13:29:40", "content": "

Thanks for creating this topic, Alessandro. We will be sure to look into any ideas offered on this thread as we prepare for the version 2 release of the Simulator.

\n

Please continue to post, everyone!

", "rawMarkdown": "", "votes": null }, { "id": "31609", "postDate": "09/26/2013 14:04:52", "content": "

Can you please release the Simulator logs of our best submission?

\n

I ask because I wasn't able to beat the sample submission (rev2). (I learned a good lesson about trying to reproduce that file early). When I run the sample file locally in the simulator, I'd get around a $12,504 mean, but I never go close to that on the leader-board. This was true of straight-to-destination routes.  I think I have a simple bug and I'd like avoid trying to find it by spending time comparing my generated file to the sample submission.  Thanks! 

", "rawMarkdown": "", "votes": null }, { "id": "31610", "postDate": "09/26/2013 14:10:26", "content": "

It looks like the wind information is defined by the closest RAP (NOAA) grid point, so if you select a flight waypoint right between two grid cells, then a minor deviation in the coordinate will allow you to "select" the more favourable wind for the next time step (0.01 hours).

\n

It'd make more sense to interpolate wind vectors between adjacent cells, so if a flight point is halfway between grid cells A and B, then the wind vectors from these cells could both have 50% weight.

\n

 

\n

Another issue, currently, if a flight starts its next "hop" (0.01 hours) in a cell with a favourable wind, and gets into unfavourable cell, it gets halfway through with an advantage. This could be exploited to improve the score without real benefit to the customer.

\n

I suggest to reduce the time step to, say, 0.001 hours. This will make the simulator run the test flights in 10 seconds instead of 1 second, but that is minor compared to the time taken to load the wind data from JSON.

", "rawMarkdown": "", "votes": null }, { "id": "31617", "postDate": "09/26/2013 16:33:38", "content": "

Thanks for this thread. I have 3 questions:

\n\n

 

", "rawMarkdown": "", "votes": null }, { "id": "31625", "postDate": "09/26/2013 17:51:28", "content": "

[quote=Nassim Khaled;31617]

\n

Thanks for this thread. I have 3 questions:

\n\n

[/quote]

\n

Hi Nassim,

\n
    \n
  1. Flight Quest 2 is not over, the Milestone period has ended. The competition will restart with the release of the second version of the Simulator, more training data and a new leaderboard. This will happen by Oct 11, if not sooner. Stay tuned.
    2. \n
  2. Submissions are still available, but the leaderboard will not change
    3. \n
  3.  I looked into the logs and you actually selected the submission the system used. If you don't do anything, the system selects your best submissions based on the Public Leaderboard. Because you made a selection, the system does not override your selection. I will follow-up with you on this offline.
    4. \n
", "rawMarkdown": "", "votes": null }, { "id": "31628", "postDate": "09/26/2013 21:24:46", "content": "

[quote=David Wilson;31609]

\n

Can you please release the Simulator logs of our best submission?

\n

I ask because I wasn't able to beat the sample submission (rev2). (I learned a good lesson about trying to reproduce that file early). When I run the sample file locally in the simulator, I'd get around a $12,504 mean, but I never go close to that on the leader-board. This was true of straight-to-destination routes.  I think I have a simple bug and I'd like avoid trying to find it by spending time comparing my generated file to the sample submission.  Thanks! 

\n

[/quote]

\n

Would be really usefull receive all the simulations 'Messages' for each submission somehow (eg. downloadable text file in "my submission" page?). Especially for the next phase, where there will be many things that can go wrong for a  flight and can't be re-created locally.

", "rawMarkdown": "", "votes": null }, { "id": "31630", "postDate": "09/26/2013 22:25:47", "content": "

Why/how did the leaderboard values (not people rank) change so drastically after the competition?

", "rawMarkdown": "", "votes": null }, { "id": "31646", "postDate": "09/27/2013 06:49:18", "content": "

Because out of the 14 days of predictions we had to do, 7 were used for the leaderboard up to the deadline and the other 7 for the current leaderboard (used for the prize).

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 31602, "author_name": "gabrieleseppi", "author_url": "", "post_date": "09/26/2013 12:43:47", "content": "

I second the request; having some useful information on how these files are created - at least assuming a full dataset - would provide a more complete environment to work with.

\n

Regarding scheduled flights after the cutoff, I may be wrong, but I think we have this information in flighthistory.csv. What we're missing is the actual takeoff/landing time if that happened after the cutoff. What I'm wonder is if the flights we have to compute routes for are included among those in actualLandings.csv

", "votes": null, "replies": [] }, { "id": 31607, "author_name": "joycenv", "author_url": "", "post_date": "09/26/2013 13:29:40", "content": "

Thanks for creating this topic, Alessandro. We will be sure to look into any ideas offered on this thread as we prepare for the version 2 release of the Simulator.

\n

Please continue to post, everyone!

", "votes": null, "replies": [] }, { "id": 31609, "author_name": "davidwilson0", "author_url": "", "post_date": "09/26/2013 14:04:52", "content": "

Can you please release the Simulator logs of our best submission?

\n

I ask because I wasn't able to beat the sample submission (rev2). (I learned a good lesson about trying to reproduce that file early). When I run the sample file locally in the simulator, I'd get around a $12,504 mean, but I never go close to that on the leader-board. This was true of straight-to-destination routes.  I think I have a simple bug and I'd like avoid trying to find it by spending time comparing my generated file to the sample submission.  Thanks! 

", "votes": null, "replies": [] }, { "id": 31610, "author_name": "sergeykozub", "author_url": "", "post_date": "09/26/2013 14:10:26", "content": "

It looks like the wind information is defined by the closest RAP (NOAA) grid point, so if you select a flight waypoint right between two grid cells, then a minor deviation in the coordinate will allow you to "select" the more favourable wind for the next time step (0.01 hours).

\n

It'd make more sense to interpolate wind vectors between adjacent cells, so if a flight point is halfway between grid cells A and B, then the wind vectors from these cells could both have 50% weight.

\n

 

\n

Another issue, currently, if a flight starts its next "hop" (0.01 hours) in a cell with a favourable wind, and gets into unfavourable cell, it gets halfway through with an advantage. This could be exploited to improve the score without real benefit to the customer.

\n

I suggest to reduce the time step to, say, 0.001 hours. This will make the simulator run the test flights in 10 seconds instead of 1 second, but that is minor compared to the time taken to load the wind data from JSON.

", "votes": null, "replies": [] }, { "id": 31617, "author_name": "nassimkhaled", "author_url": "", "post_date": "09/26/2013 16:33:38", "content": "

Thanks for this thread. I have 3 questions:

\n\n

 

", "votes": null, "replies": [] }, { "id": 31625, "author_name": "joycenv", "author_url": "", "post_date": "09/26/2013 17:51:28", "content": "

[quote=Nassim Khaled;31617]

\n

Thanks for this thread. I have 3 questions:

\n\n

[/quote]

\n

Hi Nassim,

\n
    \n
  1. Flight Quest 2 is not over, the Milestone period has ended. The competition will restart with the release of the second version of the Simulator, more training data and a new leaderboard. This will happen by Oct 11, if not sooner. Stay tuned.
    2. \n
  2. Submissions are still available, but the leaderboard will not change
    3. \n
  3.  I looked into the logs and you actually selected the submission the system used. If you don't do anything, the system selects your best submissions based on the Public Leaderboard. Because you made a selection, the system does not override your selection. I will follow-up with you on this offline.
    4. \n
", "votes": null, "replies": [] }, { "id": 31628, "author_name": "alzmcr", "author_url": "", "post_date": "09/26/2013 21:24:46", "content": "

[quote=David Wilson;31609]

\n

Can you please release the Simulator logs of our best submission?

\n

I ask because I wasn't able to beat the sample submission (rev2). (I learned a good lesson about trying to reproduce that file early). When I run the sample file locally in the simulator, I'd get around a $12,504 mean, but I never go close to that on the leader-board. This was true of straight-to-destination routes.  I think I have a simple bug and I'd like avoid trying to find it by spending time comparing my generated file to the sample submission.  Thanks! 

\n

[/quote]

\n

Would be really usefull receive all the simulations 'Messages' for each submission somehow (eg. downloadable text file in "my submission" page?). Especially for the next phase, where there will be many things that can go wrong for a  flight and can't be re-created locally.

", "votes": null, "replies": [] }, { "id": 31630, "author_name": "karle1", "author_url": "", "post_date": "09/26/2013 22:25:47", "content": "

Why/how did the leaderboard values (not people rank) change so drastically after the competition?

", "votes": null, "replies": [] }, { "id": 31646, "author_name": "gabrieleseppi", "author_url": "", "post_date": "09/27/2013 06:49:18", "content": "

Because out of the 14 days of predictions we had to do, 7 were used for the leaderboard up to the deadline and the other 7 for the current leaderboard (used for the prize).

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "31591": "", "31602": "", "31607": "", "31609": "", "31610": "", "31617": "", "31625": "", "31628": "", "31630": "", "31646": "" }, "source": "meta" }