march_madness

Data

• raw data found in /datasets/kaggle_data
• predictors found in /datasets/our_data
  • each predictor is a matrix with the rows being the year and the column being the team ID
• predictions for 2015 and 2016 are in /datasets/predictions

Modules

• these are helper files that allow us to easily setup infrastructure to train/test models and run tournaments

• march_madness_classes : code to setup tournaments, simulations, and ensembles

• march_madness_games : code to query the database

• march_madness_models : code to faciliate head to head models (used in combo with march_madness classes)

• march_madness_train_and_tune : code to do train test split

Data Exploration/Cleaning

• these files are used to generate predictors from the regular season data, and to explore relation with the response

• Kaggle Database : code to take our first crack of exploration

• Markov : code to model the ncaa as a markov chain/ explore connection with response

• Advanced : code to test some more predictors

Modeling

• Baseline : code to train a baseline head to head model
• Model Selection : code to choose predictors and train our final head to head model
• Optimize for tourney structure : code to induce upsets/ maximize expected score
• Ensemble : code to predict many unique brackets for submission

Intrastructure

• These files are all testing of the code written in the modules to make sure everything works correctly. They can largely be ignored.

Prediction

• Our final 2 years of prediction - no tuning was done, to see how our method works.