Game AI using supervised imitation learning
Trained a convolutional neural network to play the popular infinite runner “Temple Run 2,” by Imangi Studios.
Explored efficient network architectures, different methods of data collection and methods of combatting overfitting in order to produce a generalizable model that can generate impressive in-game scores.
Data is hand-collected, using a program created in python.
A short writeup of methods and results can be found here
Car prices data exploration
Buying a car is stressful. However, not all cars are built equally and, from personal experience, many car salesmen will not eagerly disclose these inequalities.
I investigated the used car market in the GTA area, and created a website and pdf report for my findings.
Website can be found here
Predicting video game outcomes with Random Forests
Created a classification model to predict match outcomes of the popular game, League of Legends, with ~82% accuracy.
Web scraped the popular aggregator site, LeagueOfGraphs.com, and collected data from the official Riot Games API.
Used python libraries such as numpy, pandas, matplotlib and sklearn to process and visualize the data, and to fit a model.
Used a Random Forest Classifier to create a model with 82% Accuracy. More details about exploratory data analysis, visualization and model creation can be found here
Course Review System Project
Command line tool that allows users to login, view, comment on and rate courses. Built in Java, this application follows SOLID principles and Clean Architecture, and uses software design patterns to implement many features. This was the final project for my Intro to Software Design class at U of T.
Edworthy web scraping project
A bot that looks at google search results and tells me when parks are empty, so that I can go on walks.
Uses BeautifulSoup and discord.py in python.
Google results(left) gets translated to a Bot notification(right)
House prices linear regression project
Predicting house prices in Ames, Iowa.
This was a data science challenge on Kaggle. Used data cleaning and linear regression techniques to score in the top 15 percent of participants.