InverseCostWeightedUtility

As explained in Bayesian Optimization in Action

InverseCostWeightedUtility is a BoTorch utility implementation that weights information or value by the inverse of evaluation cost.

InverseCostWeightedUtility is a BoTorch utility implementation for cost-aware optimization. It weights the utility or information from each query by the inverse of its evaluation cost, favoring the most informative result per dollar. section 9.7 section 9.3.2

In multifidelity max-value entropy search, the utility:

• Uses a supplied cost model to compare queries across fidelity levels. section 9.3.2
• Prefers inexpensive low-fidelity observations when they remain informative about the objective. section 9.3.2
• Switches toward high-fidelity observations when low-fidelity queries stop providing information or approach the same cost. section 9.3.2

Attach the utility to a multifidelity MES policy

Create the utility with `InverseCostWeightedUtility(cost_model=cost_model)`, then pass it as `cost_aware_utility` to `qMultiFidelityMaxValueEntropy` alongside a Sobol candidate set and a projection to target fidelity. section 9.3.2

Figure 9.10 The formula of the ROI acquisition score for multifidelity optimization. This score quantifies the amount of information a query provides about the objective’s optimum for each unit of cost.

The ROI acquisition-score figure visualizes information gained about the objective optimum for each unit of query cost. section 9.3.2

How Bayesian Optimization in Action covers it

How deep it goes

Defines inverse-cost weighting and its policy integration, but does not develop the utility’s internal implementation.

Examples

Uses low- and high-fidelity queries, a 1,000-point Sobol candidate set, and ROI acquisition-score visualizations.

What it assumes

Assumes a multifidelity model, a cost model, candidate points, and a projection from queried fidelity to target fidelity.

What you can do after

Apply it via BoTorch by supplying a cost model to a multifidelity max-value entropy search policy.

How it connects

Connects cost models, multifidelity optimization, max-value entropy search, Sobol sequences, and target-fidelity projection.

liveBookmanning.com

Read Bayesian Optimization in Action on liveBook Get the book at manning.com

Related concepts

• utility functionimplements · A utility function maps outcomes or alternatives to numerical values representing their desirability or preference for a decision maker or objective.
• max-value entropy searchused by · Max-value Entropy Search is a Bayesian optimization acquisition policy that selects evaluations expected to reduce uncertainty about the global maximum value of an objective.

This explainer was generated by Manning's knowledge graph from the text of Bayesian Optimization in Action. The section links point to the parts of the book it draws on.