I have a trained a lot and still can't figure out how people are reaching above the 50 score. It's quite impressive! I will like to learn a lot from the winners soon.
0.50 atleast is possible without surpassing the time limit and with a single model (i think 0.51 as weelll) but anything greater than 0.52, I honestly doubt if it is from a single model and if it is from a single model then definitely not within the time limit and compute restrictions. I might be wrong though. The only way to tell is if they tell us themselves, but if they don't we are going to have to wait until the comp ends . I hope one of them is kind enough to tell us :)
I dont't even think if they tell us it will give us a competitive edge tbh(you won't know how they trained their model to reach that score really), so I don't know why they dont wanna tell us
Same here. I'm not getting 0.4 😪
Did you do what I said you should do?
What was that? I am keen to learn @CodeJoe
Oh sorry, I skipped your message. Train for longer epochs. And try some postprocessing
0.50 atleast is possible without surpassing the time limit and with a single model (i think 0.51 as weelll) but anything greater than 0.52, I honestly doubt if it is from a single model and if it is from a single model then definitely not within the time limit and compute restrictions. I might be wrong though. The only way to tell is if they tell us themselves, but if they don't we are going to have to wait until the comp ends . I hope one of them is kind enough to tell us :)
Me too. I hope so.😭
I dont't even think if they tell us it will give us a competitive edge tbh(you won't know how they trained their model to reach that score really), so I don't know why they dont wanna tell us
It is a competition I get it😭😭
What did you do by the way?
train for longer I guess, plus I think the scores fluctuates and there is a little bit of a luck aspect in finding the right split
I see. Thank you @Koleshjr
@CodeJoe @Koleshjr Just to confirm, my 0.51 score is a single model. I think, to some extend, 0.52 is reachable.
thanks @Muhamed_Tuo