Limiting Runs for Computing Probability Estimates from Computationally Intense Models
Abstract
Computing probability estimates in a complex model with stochastic logic has challenges with respect to the nature of the underlying distribution, which in our effort is assumed to be binomial. We use a highly complex and computationally intense model to estimate probabilities of multiple outcomes conditioned on engagement scenarios and using thousands or even millions of iterations. Because of the amount of computation time needed, and the increasing use of the model, limiting the number of iterations is important. From a binomial perspective we have a response range on (0,1), but our model response range includes the interval endpoints and thus is [0,1]. It is the endpoints of zero and one that provide those challenges. Using a fixed value as a requirement or a relative requirement is an oversimplified approach to a conditional problem. This report details a hybrid approach to provide a user-customizable solution.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2020
- Accession Number
- AD1106181
Entities
People
- Craig D. Andres