GroverCode : code canonicalization and clustering applied to grading

Abstract

Teachers of MOOCs need to analyze large quantities of student submissions. There are a few systems designed to provide feedback at scale. Adapting these systems for residential courses would provide a substantial benefit for instructors, as a large residential course might still have several hundred students. OverCode, one such system, clusters and canonicalizes student submissions that have been marked correct by an autograder. We present GroverCode, an expanded version of OverCode that canonicalizes incorrect student submissions as well, and includes interface features for assigning grades to submissions. GroverCode was deployed in 6.0001, an introductory Python programming course, to assist teaching staff in grading exams. Overall reactions to the system were very positive.