JojoGrad

A minimal C++ library for reverse-mode automatic differentiation.

Variables form a computational graph as you apply arithmetic operations on them. Calling backward() on any node walks the graph in reverse and accumulates partial derivatives (gradients) into every upstream variable via the chain rule.

Build

cmake -B build
cmake --build build

Requires CMake ≥ 3.14 and a C++17 compiler. Tests are fetched and built automatically via FetchContent (GoogleTest v1.14).

Run tests

cd build && ctest --output-on-failure

Run examples

./build/example_basic_ops
./build/example_chain_rule

Usage

#include "jojograd/variable.hpp"

jojo::variable<double> a(3.0);
jojo::variable<double> b(4.0);

jojo::variable<double> c = a * b;
c.backward(1.0);

// a.grad == 4.0  (dc/da = b)
// b.grad == 3.0  (dc/db = a)

Operands must outlive any result derived from them — the graph holds raw pointers.

Supported types

Explicit instantiations for double and float are provided out of the box. To use another numeric type, add this in one of your .cpp files:

template class jojo::variable<MyType>;

Supported operations

+ * - / unary -