Mixed-integer models for where to build and expand childcare facilities at minimum cost and maximum fairness.
Background
Team project for Optimization Models. A region is a childcare desert when available slots fall at or below one-half (high-demand areas) or one-third (normal-demand areas) of the population of children aged 2 weeks to 12 years. We addressed three tasks: eliminate all deserts at minimum funding; repeat with a realistic cost function and a minimum spacing of 0.06 miles between facilities; and maximize fairness by minimizing differences in supply-demand coverage across regions.
My role
Led the team: clarified requirements and deliverables, and split the work so pairs owned each sub-task.
Formulated and coded the models for tasks 2 and 3 in Python with Gurobi, including the facility-spacing distance constraint.
Cleaned and joined population, income, and existing-facility data by ZIP code with pandas and NumPy.
Created map visualizations of recommended new facility sites and expansions to existing ones.
Results
Delivered a Jupyter notebook and a report with models, results, and recommendations for the New York State government.
Translating policy definitions into precise constraints is most of the work in a real optimization problem.