Ikr, If people are getting good results by postprocessing I rather stay out of it because I know what post processing did to me in that challenge😂😂

😂😂😂I won't forget

🤣such a painful experience

That challenge has taught me to trust my approach always 😂😂😂

I put all my bets on @ihar

😂😂

why not on yourself yan?

I have a feeling that I'm the only one in the top 10 using a classical ML approach, so I might be wrong.

Where is my friend @JuliusFx? Please bless us with your wisdom.

All in all. This competition has been fun. I might have struggled but I have had so much fun working on this project. And I cannot wait to learn from you guys when the competition ends. This one deserved a better prize for sure.

Hello my good friend @yanteixeira! Tried RNNs based models. They take long to train and results not yet decent..So might just stick with ML. Good job you are doing there on the leaderboard.

There is a possibility of a huge shakeup when the competition closes. I don't have very decent local RMSE! personally. So my best models on leaderboard are the worst on cross eval. My best on cross eval the worst on LB. Disturbing!

Yeah, I think I know exactly what is causing this, but unfortunately, I can't share it right now. Let's just say that our models can't predict what they have never been shown ;)

Well understood, kindly do share your insights when the competition closes. All the best

I have the same experience. There is almost no correlation between my CV and LB scores. The approach just submit different solutions to minimisze the LB score is the clear way into total overfitting.

Ego is the name of the model that minimize the LB