UNICEF Arm 2030 Vision #1: Flood Prediction in Malawi
'/%3e%3c/g%3e%3cuse xlink:href='%23b' transform='rotate(10)'/%3e%3c/g%3e%3cuse xlink:href='%23c' transform='rotate(20)'/%3e%3c/g%3e%3cuse xlink:href='%23d' transform='rotate(40)'/%3e%3c/g%3e%3cuse xlink:href='%23e' transform='rotate(-80)'/%3e%3c/g%3e%3cpath d='M-225-21.5h450v7.5h-450z'/%3e%3cpath fill='%23339E35' d='M-225 86h450v100h-450z'/%3e%3c/svg%3e)
Hello Zindians,
Again the Team from Ukrania are using the raster that they found with some manipulations and that's obvious they were 200+ on the leaderboard jumped straight to the top15 and then from 0.075 score directly to 0.068 and I'm sure that they can do more It's just a sort of camouflage.. Since the first foul that they did Zindi had to disqualify them . That's unfair with a 0.028 score they can know the distribution of the target and they can use that knowledge in a tricky way in their model .. C'mon they know the target!!!! What are we doing here ?! Even for the Unicef the model will be useless .That's not Data Science .. We have been working on the problem during months to build better models for the sake of humanity first of all and we all know such models won't help .. I hope that Zindi understands that .
Cheers
We can classify that additional data as a leak , using that will lead to a model that do not generalise .. it will lead to overfitting on another set of data !
Fadhloun maybe you should trust Zindi that they will do whatever it takes to make the competition fair? If they review the submission and found that it's influenced by the target they will disqualify them. Same with yours. And with everyone else's. Just do the best you can and leave the rest to the company that runs this show.
@RenierBotha, the problem is that once you know the answers you can manipulate your solutions to get as far or as close to the RMSE as you want. One example I can point out is use early stopping with the actual flood 2019 as the predictions. then you can just remove the original target from the solution, and we can never find out. Its very easy to remove the traces of being influenced by target from your code, so I guess its not a good parameter to judge the authencity of solutions. I am not blaming anyone, but I am saying that its very easy get a good score in this case. And the team doesn't need to wait for the private leaderboard, they already know their score :P
Hey :D
Yeah, thats a very good point and something Zindi will have to take into account.
What I am alluding to, though, is that it is as much (indeed more) in the interest of Zindi to ensure the chosen winning solution(s) are useful and fairly chosen. If someone wins unfairly, they get lucky. But if Zindi chooses an undeserving winner, they lose. Big time.
So my point is, instead of making emphatical discussion posts accusing others of cheating, my opinion is that it is best to leave the judgement to the judges.
Totally agree with you Renier, may the best submission win :D