Hi 👋, I've implemented the Damped Lagrangian Formulation to enhance the stability of the optimization process by addressing oscillatory behaviors when constraints are suddenly satisfied or violated. This involves:
Adding the DampedLagrangianFormulation class which extends the traditional LagrangianFormulation with a damping mechanism.
Updating the documentation in lagrangian_formulation.rst to describe the new formulation and its implementation details.
Testing
Added tests in test_lagrangian_formulation.py to ensure:
The damping effect is properly computed as detailed in this blogpost.
References
John C. Platt and Alan Barr, "Constrained differential optimization," presented at Neural Information Processing Systems, 1987.
Engraved Blog, "How We Can Make Machine Learning Algorithms Tunable," 2024. Available online: Engraved Blog
Changes
Hi 👋, I've implemented the Damped Lagrangian Formulation to enhance the stability of the optimization process by addressing oscillatory behaviors when constraints are suddenly satisfied or violated. This involves:
DampedLagrangianFormulationclass which extends the traditionalLagrangianFormulationwith a damping mechanism.lagrangian_formulation.rstto describe the new formulation and its implementation details.Testing
Added tests in
test_lagrangian_formulation.pyto ensure:References