A function evaluation, also called a query, computes an objective function’s output for a specified input. In black-box optimization, each trial tests a particular setting and observes its result, often at substantial time or computational cost. section 1.1.2
Function evaluations determine how an optimization loop uses available resources:
- Sequential optimization makes one query, receives its function evaluation, and then selects the next query. section 7.0
- Batch optimization makes multiple queries in parallel, increasing experimental throughput when processing units or physical resources can be parallelized. section 7.0
- Selecting several nearby high-scoring points can waste evaluations by clustering experiments around essentially one input location. section 7.1.2
Airplane design optimization
A simulated airplane objective is evaluated at points in a four-dimensional unit hypercube. An experiment can use 100 total function evaluations, arranged as batches of five, and record the best value found as optimization proceeds. section 7.5
Figure 1.2 The framework of a black box optimization problem. We repeatedly query the function values at various locations to find the global optimum.
The diagram shows the repeated-query framework: evaluate objective-function values at different locations while searching for the global optimum. section 1.1.2
