Really want to know how this was achieved. My bet is also on leaks being found, given the way the top two teams dramatically improved their score hours from the deadline
You are absolutely right. That or they pretrained on previous competition's data (which should be from the same source). The real leaderboard starts from #5.
Really want to know how this was achieved. My bet is also on leaks being found, given the way the top two teams dramatically improved their score hours from the deadline
My bet it is using this dataset https://mlhub.earth/data/lacuna_fund_eotg_v1
(It is linked in the competition description)
I think it is a leak score.
Just for clarification, is this an external dataset (https://mlhub.earth/data/lacuna_fund_eotg_v1)? if yes, is the use of external open dataset allowed ?
You are absolutely right. That or they pretrained on previous competition's data (which should be from the same source). The real leaderboard starts from #5.
Yes, no. Afaik
Yes. You should.