CS74: Predicting basketball statistics

Dependencies

• MATLAB R2014a
• scrapy
  • Run pip install scrapy to install

Data

• All data is contained in the data directory. The data was all collected from basketball-reference.com. We used total and per-possession data from the 1979-80 season, when the three-point line was introduced, to the present. column_headers gives a short description of each column of information.
• To scrape your own data, you must first set up the python environment by running in the home directory source env/bin/activate. Then run scrapy crawl curry in the scraper/scraper directory to crawl basketball-reference.com for data.

MARS

• To run pre-selected MARS regressions, call run_mars from the mars directory. This script tests the MARS algorithm on per-possession data to predict two-point percentage, assists, total rebounds, and points for various positions.
• Call interface in the mars directory and follow the given directions to test MARS on parameters of your choosing.

GBDT

• To see how GBDT works, call run_gbdt from the gbdt directory. This script will run the GBDT algorithm on per-possession data and predict two-point percentage.

External Code

• All files in the env directory are from virtualenv. All files in the scraper directory, with the exception of items.py and bballspider.py, are from scrapy.