For our class project, we developed a hero recommendation engine for the popular computer game Dota 2. We used Valve's Steam Web API to aggregate data for over 50,000 matches. Using logistic regression and k-nearest neighbors models, we were able to predict match outcomes with 69% accuracy based on the teams' hero compositions alone. Finally, we used k-fold cross validation to determine an optimal weighting function for the k-nearest neighbors model.
Stanford Advanced Topics in Networking with Nick McKeown and Sachin Katti
Collaborators: Sandeep Chinchali
For this project we experimentally determined a tighter bound for Data Center TCP's (DCTCP) marking threshold. To do this, we used Mininet to create a star topology with two senders. Given varying round trip times, link capacities, and marking thresholds, we were able to determine the robustness of the theoretical marking threshold given in the original DCTCP paper.