voiage.ml_policy_voi
ML, LLM, and Agent Value of Information: Decision-Focused Model Value & Policy Uplift VOI (#576, #578).
This module implements:
- Decision-focused model evaluation, validation, and refresh triggers (#576): comparing models by downstream decision loss and policy regret rather than pure accuracy or AUC.
- Policy and uplift value of information (#578): valuing heterogeneous treatment effects (CATE), potential outcomes, intervention costs, and constrained targeting.
Classes
Section titled “Classes”- CandidateModelEvaluation Evaluation summary for a single candidate ML model.
- DecisionFocusedModelValueResult Result of decision-focused model value evaluation and refresh analysis.
- UpliftVOIResult Result of heterogeneous policy and uplift VOI calculation.
Functions
Section titled “Functions”- evaluate_decision_focused_model_value Evaluate candidate models by downstream economic decision value.
- compute_policy_uplift_voi Compute heterogeneous policy and uplift Value of Information.
- validate_decision_focused_model_value Validate a dictionary against the decision-focused model value schema.
- validate_policy_uplift_voi Validate a dictionary against the policy uplift VOI schema.