Prototype implementation of automatic differentiation for learning purposes while reading Evaluating Derivatives: Principles & Techniques of Algorithmic Differentiation (2nd Edition). Starting with forward mode, and (if time permits) moving on to reverse mode.
Currently working functionality is shown in the unit tests:
- Forward mode differentiation for scalar-valued functions utilizing a subset of operators
- Gradients for vector-valued functions
- Row-wise and columnar Jacobian calculations for vector to vector functions
Built without LLM assistance; references linked where appropriate.