erikcs.r-universe.dev

[erikcs] balnet 0.0.4

erik.sverdrup@gmail.com (Erik Sverdrup) — Sun, 23 Aug 2026 10:10:23 GMT

Provides pathwise estimation of regularized logistic propensity score models using covariate balancing loss functions rather than maximum likelihood. Regularization paths are fit via the 'adelie' elastic-net solver with a 'glmnet'-like interface, yielding balancing weights that target covariate balance for the ATE and ATT. Under lasso penalization, lambda bounds the maximum covariate imbalance, so the regularization path traces a sequence of decreasing imbalance tolerances. For details, see Sverdrup & Hastie (2026) .

[erikcs] maq 0.6.1

erik.sverdrup@gmail.com (Erik Sverdrup) — Tue, 04 Aug 2026 13:10:17 GMT

Fits generalized Qini curves for evaluating treatment allocation policies with one or more mutually exclusive treatment arms. Given estimated conditional treatment effects and treatment costs, 'maq' computes the optimal treatment allocation over a range of budget levels and estimates the resulting policy value using doubly robust or inverse-propensity weighted scores. The underlying optimization exploits the linear multiple-choice knapsack structure of the allocation problem to compute the full solution path efficiently. For details see Sverdrup, Wu, Athey, and Wager (2025) .

[erikcs] policytree 1.2.5

erik.sverdrup@gmail.com (Erik Sverdrup) — Tue, 04 Aug 2026 10:30:29 GMT

Learn optimal policies via doubly robust empirical welfare maximization over trees. Given reward estimates, the algorithm finds a rule-based treatment allocation, where the policy takes the form of a shallow decision tree that is globally optimal (or nearly so). Methods are described in Sverdrup, Kanodia, Zhou, Athey, and Wager (2020) , Athey and Wager (2021) , and Zhou, Athey, and Wager (2023) .