README

The program is run through predictor.py. Type in python predictor.py to run with default options (default team is the Niners, algorithm is SGD w/heuristics). Training data is contained in the two gamedata txt files; validation data is in the two validation txt files.

If you want to run the program to get the results mentioned in my appendix, run the following commands (these run through all the relevant predictor arguments):

python predictor.py -t PHI -p basic
python predictor.py -t SFO -p basic
python predictor.py -t PHI -p bayes
python predictor.py -t SFO -p bayes
python predictor.py -t PHI -p bayes -z 1
python predictor.py -t SFO - p bayes -z 1
python predictor.py -t PHI
python predictor.py -t SFO
python predictor.py -t PHI -p pivot
python predictor.py -t SFO -p pivot
python predictor.py -t PHI -p pivot -a ava
python predictor.py -t SFO -p pivot -a ava

Explanation of relevant options:
-p: changes algorithm (basic = predict most common play, bayes = naive bayes classifier, sgd = SGD with heuristics, pivot = multiple binary classifiers)
-t: changes team (use SFO or PHI)
-z: changes Laplace smoothing for Naive Bayes classifier
-a: changes approach used for multiple binary classifiers (ova = one vs. all, ava = all vs. all)
There's also a -1 option that predicts a single play using user input and SGD w/heuristics (enter -1 yes or -1 no, default is no)

Since I adapted the baseline SGD code from homework 7, there are also loss, initial step size, step size reduction, regularization, and verbosity options applicable to the SGD process.