As this competition closes, I hope Zindi is keeping an eye on the LB to fish and investigate any suspicious activity.
I woke up only to find massive leaps in the last few hours. Fellow Zindians, from your honest smart work on this challenge, please what is the likelihood of gaining 0.003 (0.07... Range) improvement in 24hours. Please comment your thoughts below. I am looking forward to reading your opinions.
I am pretty sure, there's something fishy, in this. I am a huge fan of magic, but this is something of other universe 😂. Totally agree with you Dr
😂😂😂 @ huge fan of magic
cheating of any form erodes the awesomeness of a community,I hope that is not happening here.As the competion comes to an end ,Zindi should extend its stance particulary on the challenge!
I hope so too man. All the best as it comes to an end.
@Nikil very fishy indeed.. Can’t just keep laughing.. high cheating has began for the past 4weeks.. I just calm down to see what happens after the challenge ends. Don’t be surprised who will also be caught as the challenge comes to an end. I observed every single drastic change on the leaderboard which is so obscure.. I warmly await @zindi code review.
Yes Engineer, very sad indeed, I wont be surprised at all, I know :D
Since past weeks this challenge has been full of fishy things, it got to a point I got tired of working on it, I know how hard it takes to improve ones performance please let's not set a bad record here, zindi has been a very good platform for learning and I started my competitive data science here. It will be a bad record for zindi if cheating is allowed here. Good luck to everyone.
I can relate to what you are saying. It affected my team in the ICLR challenge where lots of fishy things were happening without being addressed so I gave up too soon and it affected overall ranking. My advice is to push till the last second and hope Zindi detects the cheating activity. All the best.
Very likely, due to change in model architecture; my score changed by 0.01 days back as a result of this... 0.003 isn't so huge as a 0.1 would be at this point
True, my score improved at some point by even more that 0.1. But I actually meant in the 0.07 range, as you get higher it becomes more difficult to make that leap.
Still, its very unlikely that you or other teams are using the same features or even models and hyperparameters as the 'magicians' in question. So you can't rule out the .003 improvement..
Fair point. Not ruling out the possibility. But from my experience with this competition that scenario will be extremely rare at this stage.
Yes @DrFad.... I underlined that phrase " extremely rare at this stage" . Yes I agree totally that scenario of such possibility is too rare indeed. I could remember that I started noticing this sudden and strange lift in this LB. I just kept calm since then and allow you guys to see that by yourselves. Here now, we can all noticing that.
I am new to the field. I am following with keen interest.
I think here you are talking about the Team be careful and when you talk about it you should be careful :D (just kidding) .
There is no magic, we just come back and our improvement with 0.003 is just the fruit of strong data analysis and features engineering and well understanding for the problem, btw no external data used.
I will be happy to share my solution when the review ends to learn from it and you will see how simple and straightforward our solution is, I am sure everyone could do this if they focused on the data itself instead of focusing on looking for external data.
Next time please focus on your game and keep others working in peace otherwise this competition will be my last one because I am tired and fed up from this kind of discussion in each competition I am playing on it.
Hi Mohamed. You spoke well until your last paragraph. That wasn't necessary.
No one is questioning whatever feature engineering or analysis you have put in your code. The point in hand is how genuine you were in getting them without help from a team or competitor. This suspicion is benchmarked based on the significant jump at that score range i. e. Circa 0.07... Which is clearly an outlier from the 480 data scientist who made submissions to this challenge. I have been watching the LB for months, for this challenge, everyone within the top 10 range moved gradually to earn that spot. This is a true reflection of dedicated time and obtained results, hence, the suspicion with your teams case because it was an anomaly.
My team didn't use external data and I am sure a lot of people didn't as well, so you can set aside your statement on external data.
Indeed, if you didn't solicit help from a participant to make that questionable leap, then welldone - kudos to you and your team mate. Otherwise, I hope Zindi preserves the integrity of this platform and reward the right entries.
"""I have been watching the LB for months, for this challenge, everyone within the top 10 range moved gradually to earn that spot.""" Have you watched the first team move? They were 200+ on the leaderboard and jumped directly to the 1ST Position again after Zindi removed there submission with that leak (0.028 RMSE) and None of you reacted even if that's obvious .
There wasn't a need to react. The person in question immediately opened up and detailed the steps he took to notify Zindi. He even took time to explain why... That's honesty. It was an honest mistake while exploring data sets. Or would you rather we tell Zindi what they already know?
Going up again to number 1 will be subject to code review.
Well first that's a big unforgivable mistake that we cannot pass away .They had almost the test target since the score was so close to zero.How did you know that they didn't use any unallowed external data ? Again I'm talking about the second time that they jumped to the 1ST position .. From 200+ to 1ST in two days and from 0.075 to 0.068 just after that fault and Again you didn't react !!! If you trusted them why would you ask them about the raster that they used in the discussions?
Aaah Team BeCareful won't be subject to code review??
@Drfad .. i totally disagree with such improvement under 24hrs (at the closing hour of the challeng). ‘something is fishy’ and Zindi must take necessary steps.. it’s not time to to voice out completely; I quietly wait to see what the final leaderboard will look like..