GAMS launches GAMSPy tools to merge optimization and ML for grid modeling

Adam Christensen of GAMS detailed GAMSPy, a Pythonic modeling layer that brings GAMS algebraic modeling together with common ML toolchains (PyTorch, NumPy) and commercial solvers. He described new formulation libraries (activation functions, linear layers, surrogate model embedding) that let engineers embed trained neural‑net layers and surrogate models into optimization formulations for use with solvers such as CPLEX and Gurobi. "GAMSPy lets you formulate models in Python and hand them to best‑in‑class solvers without leaving the Python environment," Christensen said.

Why it matters: Optimization and ML communities increasingly intersect for applications like market simulation, planning and surrogate model embedding. Christensen emphasized reproducibility and academic access: GAMS announced free academic licenses for GAMSPy integrations that include access to commercial solvers and an academic program to enable wider testing and teaching.

Details: The presentation covered syntax, container‑based model objects, dense and sparse data handling, formulation generators for common ML activations (ReLU, softmax), and example workflows to integrate PyTorch trained layers into GAMS optimization. Christensen noted early error detection, domain violation checks and a library of model pairs (classic vs GAMSPy) to assist migration.

Next steps: Christensen encouraged attendees to test GAMSPy, consult online documentation, and contact GAMS for collaborations; he noted research and product roadmap items such as expanding supported ML model types and training in the GAMS environment.