Details

A Python autograder which uses supervised ML and basic NLP techniques to grade student code reading responses for CS 105, an intro-level CS course at UIUC. It uses annotated code reading responses from previous CS 105 semesters. Binglin Chen built the initial autograder models, and Max Fowler was responsible for initially acquiring, grading, and transforming student responses for the ML models. My role was to add recently collected responses to the dataset, experiment with different classifiers and feature sets, and to apply basic NLP techniques to the responses to improve accuracy, all of which successfully improved the autograder by about 1% across all question sets. It currently is comparably accurate to course TAs, and is used in the class today.