Welcome to the 2025 INFORMS Annual Meeting at the Georgia World Congress Center and the Omni Atlanta Hotel at Centennial Park in Atlanta, Georgia, where more than 6,000 INFORMS members, students, prospective employers and employees, and academic and industry experts will share the ways O.R. and analytics are fueling Smarter Decisions for a Better World.
Sign up for our general information newsletter to stay up-to-date!
Presented by: Adam Christensen
Recent advances in ML/AI have commoditized the development of surrogate models using tools such as PyTorch, Scikit-Learn, and TensorFlow. These surrogate models simplify inherently non-linear phenomena, approximating complex behaviors so they can serve as constraints in optimization frameworks. Embedding these models in algebraic modeling languages (AMLs) like GAMS remains challenging: designed for sparse algebra, AMLs lack seamless integration with third-party software. The rise of Python in data science has motivated a paradigm shift, inspiring tools that bridge classical AMLs and current computational techniques. We introduce GAMSPy, a native Python AML combining the mathematical transparency and scalability of traditional AMLs with Python’s ecosystem. Its set-driven constructs and operator overloading preserve the syntax of handwritten algebra while supporting dense matrix operations—matrix multiplication, transposition, norms—essential to ML/AI. While the GAMS “classic” engine excels at indexed algebra, GAMSPy extends its capabilities to accommodate ML workflows. We demonstrate embedding a neural network trained in PyTorch to model an energy system as a constraint within an optimization problem, enabling system engineers to optimize plant operations with detailed energy conversion models. This workflow exemplifies applications spanning weather forecasting and market behavior modeling. We also compare GAMSPy to existing approaches, discuss future developments, and highlight innovative intersections of mathematical modeling and machine learning. GAMSPy represents a significant convergence of AML rigor and Python-driven ML versatility. Its design prioritizes computational efficiency, syntactic clarity, and scalability, offering a robust platform that overcomes integration hurdles and unlocks new possibilities at the intersection of optimization and data science.
