A C++ optimizer buffet served header-only

What it does ensmallen is a header-only C++ library for non-linear numerical optimization. It bundles a wide range of optimizers—covering gradient descent, gradient-free methods such as CMAES and Simulated Annealing, and constrained techniques—behind a unified interface. You provide an objective function and select an algorithm.

The interesting bit Because the entire library ships as headers, there is no separate compilation or linking stage to fight with. It also exposes optional callbacks, so you can hook into the optimization loop for logging or early stopping without modifying the underlying solvers.

Key highlights

• Ships L-BFGS, SGD, CMAES, Simulated Annealing, and other constrained and unconstrained methods.
• Header-only distribution; drops into existing C++ projects without library linking.
• Optional callbacks allow customizing the optimization process mid-run.
• Backed by a 2021 JMLR paper and the broader mlpack ecosystem.
• Requires only C++14, Armadillo, and a standard BLAS/LAPACK backend.

Verdict Grab it if you need flexible, standalone function minimization in C++ without adopting a monolithic framework. Pass if you are already locked into a larger C++ ML stack that manages its own optimizers.