Generate fixes for incorrect python code, given data about the problem being solved
A good entry point is end_to_end.py: it's an end-to-end example of how the system is designed to work.
Given:
- An incorrect solution to a programming problem (usually written by a student)
- Some set of correct solutions (often written by other students)
- Some set of correctness checks (unit tests)
Try to generate a (new) corrected solution that is close to the incorrect solution, but fixes all the bugs.
Present this correction as a series of fixes, and provide explanation about how this fix improves things: how does it change what the code does when it runs?