Policy and Uplift VOI

compute_policy_uplift_voi values personalized decision policies driven by heterogeneous treatment effect models (e.g. Causal Forests, Meta-learners, EconML) under finite intervention budgets.

Features

• Computes optimal targeted allocation under hard budget limits.
• Evaluates clairvoyant Uplift EVPI across posterior CATE trajectories.
• Decomposes uncertainty into subgroup-specific EVPPI for targeted data gathering.

import numpy as np
from voiage.ml_policy_voi import compute_policy_uplift_voi

# 100 simulations across 500 customers
np.random.seed(42)
cate_draws = np.random.normal(loc=0.08, scale=0.04, size=(100, 500))

subgroups = {
    "enterprise_tier": np.array([True] * 100 + [False] * 400),
    "smb_tier": np.array([False] * 100 + [True] * 400),
}

uplift_res = compute_policy_uplift_voi(
    cate_samples=cate_draws,
    intervention_cost=50.0,
    payoff_multiplier=1200.0,
    budget_constraint=5000.0,
    subgroups=subgroups,
)

print(f"Targeted Units: {uplift_res.units_targeted} / 500")
print(f"Optimal Policy Net Value: ${uplift_res.optimal_policy_value:,.2f}")
print(f"Uplift EVPI: ${uplift_res.uplift_evpi:,.2f}")
print(f"Enterprise Subgroup EVPPI: ${uplift_res.subgroup_evppi['enterprise_tier']:,.2f}")