Automatic Differentiation Engine
For my final project in Harvard’s CS107 Systems Development course, I wrote a Python package for automatic differentiation called graddog. It computes derivatives of numerical functions using both forward and reverse mode autodiff.
For extra credit, I implemented automatic second-order differentiation to calculate Hessians via a method called Edge Pushing. The idea is to propagate second-order derivative information through the computational graph by “pushing” partial derivatives along edges.
