slides.pptx
- 1. Trying to build off of last summer’s work!
- 2. Last Summer • Allocative optimization for South Africa model. • 3 parameter problem – ART, PrEP, VMMC • Solved through random sampling of the parameter space, manual estimation of trends + a little bit of statistical analysis • Goals: • Find a more efficient/organized solution • Generalize approach to higher dimensions/other models.
- 3. Generated Hypercube in Parameter Space • 11 x 11 x 10 points sampled - DALYs and Cost calculated for each point. • Built surrogate model for Objective (DALYs) and Constraint (Cost) • Model represents DALY, Cost as a product of univariate functions in ART, PrEP, and VMMC. (Separation of variables) • Used surrogate models to solve constrained optimization problem.
- 6. Findings • ART has the most influence on DALYs. • ART, PrEP have around equal influence on cost. • Main strategy: invest in ART, remainder goes to PrEP • Suspicious observation: • VMMC parameter has almost no effect – did I make a mistake? (I probably made a mistake)
- 7. Next Week • Future work – I found additional parameters in the model: • 8 for ART • 1 for PrEP • 1 for VMMC • 1 for HCT • Generating 1000 random points in the 11-dimension hypercube. • Generate surrogate model using separation of variables. • Use the same code to solve the constrained optimization.