I decided to try some interesting things in this competition , and managed to reach 16th in LB (now 22) using a non ML solution.
If any of you up there in top 10 want to team up in order to use the data as features or simply combine\blend with a different kind of approach ,
please feel free to contact me on email@example.com
My teammate and I are very interested in your features. Would you mind sharing some idea about how you create your features. Looking forward to your reply.
I don’t have any “features” per se, as I don’t use any ML algorithm.
I did do some extensive alteration to the data in PowerBI and wrote some R scripts with some pretty deep logic to generate predictions.
I think that implementing calculations/features + the actual predictions from within my solution into an existing LSTM that I assume you guys are using, will be very powerful. Even only blending with the LSTM prediction might prove useful as the 2 methods are very different.
Additionaly, I didn’t use temperature data at all.
And there are more improvements to be done.
Sorry for not sharing more, if I’m one of a few, or the only one to use this kind of non ML solution, this should be very powerful and I’m looking to join a top team to utilize this.
If you are interested in adding me to the team, that will be great, we can Skype to talk further before committing.
Just to note, for anyone reading this, i know it’s against the rules to share information with competitors from other teams, and i do not intend to share privately any such information before actually joining a team.