March Madness ML training, from kaggle, march machine learning mania
I used a virtual env for this, you should do the same so our version match.
python -m venv ballenvy
source ballenvy/bin/activate
pip install .
Data is held at the kaggle URL. You must log in to download the folder. Set the location of your data folder in the config file. I couldn't be bothered to configure the kaggle package.
I'd like to submit a few different predictions:
- A submission that is all random numbers (random choice) - DONE (Random_values)
- A submission that is all .5 (no choice) - DONE (Equal_values)
- My own ELO calculated from the bare minimum - DONE (BasicEloProbs)
- My own ELO calculated from Nate's methodology - DONE (ComplexEloProbs_noconfreversion)
- another file with conference mean reversion - DONE (ComplexEloProbs)
- Nate's ELO on its own - DONE (NateEloProbs.csv)
- Nate's ELO + ????, decide what to add to move it above EV
- Composite with other power rankings?
- apply a 1.07x boost to elo differences for tourney play - Done (to all...)
- Feature Engineering from historical data, using elo paired with seed difference, recent performance?
- I should use one that does it based on ratings alone
- A composite of mine + Nates
- A version of the composite with select teams probability set to 100% (1,2,3 seeds)
- Re-run them all after selection sunday
All of these will need to get updated when the newest data release comes