mini_opt

mini_opt is a small C++ implementation of constrained non-linear least squares. I implemented this for fun, and use it to solve toy problems that interest me. The solver supports:

• Non-linear cost functions and equality constraints.
• Box inequality constraints via interior-point method.
• Line search using polynomial approximation or armijo backtracking.
• Levenberg marquardt.

Building

To build and run tests:

git submodule update --init
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=RelWithDebInfo
cmake --build .
ctest

Optional serialization of optimization outputs (using nlohmann) is enabled by passing -DMINI_OPT_SERIALIZATION=ON.

References

The reference used for this project is:

"Numerical Optimization, Second Edition", Jorge Nocedal and Stephen J. Wright

The implementation is mostly based on Chapters 18 and 19 of this book.

TODOs:

• Add support for non-linear inequality constraints. Only diagonal box constraints are supported at the moment.
• Add sparse versions of the solvers. For now I only support dense problems.

---

Project is licensed under MIT License.