Coulda, Woulda, Shoulda - March Machine Learning Mania 2016

The March Madness competitions have the perfect trifecta for regret: repeated leaderboard feedback, ability to tweak probabilities based on personal biases, and a wealth of different data points. So what would you do differently for next year? The only tweak I did was changing the round 1 probabilities for seed 1s to 100% on one submission. I should have gambled a lot more because my two submissions are so similar. I didn't notice this cause I just saw that the predicted champions were different without realizing all the probabilities are within 2-3% of each other.