I just placed #35 in a competition with 934 teams.
OVERFITTING!
The public leaderboard can throw off your intuition as it contains only a few samples! Trust your CV and methodology.
There were a lot of concerns over score probing in the discussion, however, it seems that the cheaters all ended up overfitting to the x% of the total test set.
You can learn a lot!
I learnt so much from this competition
- learnt there is such a thing as CatBoost
- I learnt about hyper parameter optimisation using optimise your optimisation libraries :)
- I learnt about LDA from @oscarm524
- I learnt about feature engineering from @belati
- I learnt about feature selection
It’s a lengthy but highly rewarding process
Community
The community over at Kaggle is super helpful! You can learn from discussion boards, other people’s notebooks, etc.
Every few clicks I end up seeing super unique ideas that completely blow me away!
