kwinkunks
commented
7 years ago Hey Josh... this worked out pretty well and scored 0.574.
I'm happy to hear that you've learned a lot in the process of building your entries. I am also on a steep learning curve! Like: there are a lot of entries in the last week so maybe don't score everything 'by hand'! Still, I have it down to an art now :)
Sorry you wound up in Python after all --- I have enjoyed seeing other languages in the mix. You'll have to check what Thanish was doing with F1 to get some hints about that. Anyway, the Python seems to be working out at least! Nice that you have the option of switching.
jpoirier011_submission001.csv using random forest in python - fun learning scikit-learn!
I suspect my R submission may have been using an incorrect "F1-score" - R doesn't have a built in F1-score that I'm aware of so I copied another implementation of it. If it was different than the one we're using for the contest - it could lead my algorithms astray during training/tuning.
My Python submission here follows mostly the same logic of feature engineering, tuning, prediction . . . although I came up with the idea of finding the averages for each log for each facies in the training data (what is the GR average in SS) - and taking the difference from the actual log (and squaring it to remove directionality). This added a lot of features (and perhaps bias) - I'm happy to hear any thoughts although I'm sure y'all are busy in these last couple days!
Thanks for hosting this contest - I've had a lot of fun learning!