Sculpture Style Classifier

  • Web App in which a user can upload an image of a sculpture and have it classified by style.
  • Returns a bar chart which contains the probabilities of the sculpture belonging to each class.
  • The model was built by fine-tuning the CNN Xception and was done as part of my Senior Thesis.

OffsideReview.com

  • OffsideReview is a website for modern National Hockey League statistics from the 2016-2017 season onwards for all Regular Season and Playoff games.
  • Can query statistics for skaters, goalies, and teams with a variety of different filters (e.g. date, player, team).
  • Each query makes an Ajax request to the server to retrieve the requested data.
  • At 10 a.m. every morning the new games are scraped and then processed with the new data being added to the database.

hockey_scraper

  • Designed to allow one to scrape the Play-By-Play and Time-On-Ice tables off of the National Hockey League (NHL) API and website.
  • Can scrape the information for any preseason, regular season, and playoff game from the 2007-2008 season onwards.
  • Has functionality to scrape games by season, between a given date range, and by individual games.
  • Available as a python package on pip as “hockey_scraper”.

NHL Expected Goals Model

  • Utilized machine learning techniques to create a model that predicts the probability of an unblocked shot of being a goal in the National Hockey League.
  • Three different machine-learning classifiers were used to fit the same data and features, including: Logistic regression, Random Forest, and Gradient Boosting.
  • All three models were evaluated using both log loss and by plotting it’s ROC curve and calculating its AUC score.
  • The Gradient Boosting model performed best in both evaluation metrics followed by the Random Forest model and then the Logistic regression model.