Efficient Policy Learning from Surrogate-Loss Classification Reductions

Andrew Bennett, Nathan Kallus
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:788-798, 2020.

Abstract

Recent work on policy learning from observational data has highlighted the importance of efficient policy evaluation and has proposed reductions to weighted (cost-sensitive) classification. But, efficient policy evaluation need not yield efficient estimation of policy parameters. We consider the estimation problem given by a weighted surrogate-loss classification with any score function, either direct, inverse-propensity-weighted, or doubly robust. We show that, under a correct specification assumption, the weighted classification formulation need not be efficient for policy parameters. We draw a contrast to actual (possibly weighted) binary classification, where correct specification implies a parametric model, while for policy learning it only implies a semi-parametric model. In light of this, we instead propose an estimation approach based on generalized method of moments, which is efficient for the policy parameters. We propose a particular method based on recent developments on solving moment problems using neural networks and demonstrate the efficiency and regret benefits of this method empirically.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-bennett20a, title = {Efficient Policy Learning from Surrogate-Loss Classification Reductions}, author = {Bennett, Andrew and Kallus, Nathan}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {788--798}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/bennett20a/bennett20a.pdf}, url = {https://proceedings.mlr.press/v119/bennett20a.html}, abstract = {Recent work on policy learning from observational data has highlighted the importance of efficient policy evaluation and has proposed reductions to weighted (cost-sensitive) classification. But, efficient policy evaluation need not yield efficient estimation of policy parameters. We consider the estimation problem given by a weighted surrogate-loss classification with any score function, either direct, inverse-propensity-weighted, or doubly robust. We show that, under a correct specification assumption, the weighted classification formulation need not be efficient for policy parameters. We draw a contrast to actual (possibly weighted) binary classification, where correct specification implies a parametric model, while for policy learning it only implies a semi-parametric model. In light of this, we instead propose an estimation approach based on generalized method of moments, which is efficient for the policy parameters. We propose a particular method based on recent developments on solving moment problems using neural networks and demonstrate the efficiency and regret benefits of this method empirically.} }
Endnote
%0 Conference Paper %T Efficient Policy Learning from Surrogate-Loss Classification Reductions %A Andrew Bennett %A Nathan Kallus %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-bennett20a %I PMLR %P 788--798 %U https://proceedings.mlr.press/v119/bennett20a.html %V 119 %X Recent work on policy learning from observational data has highlighted the importance of efficient policy evaluation and has proposed reductions to weighted (cost-sensitive) classification. But, efficient policy evaluation need not yield efficient estimation of policy parameters. We consider the estimation problem given by a weighted surrogate-loss classification with any score function, either direct, inverse-propensity-weighted, or doubly robust. We show that, under a correct specification assumption, the weighted classification formulation need not be efficient for policy parameters. We draw a contrast to actual (possibly weighted) binary classification, where correct specification implies a parametric model, while for policy learning it only implies a semi-parametric model. In light of this, we instead propose an estimation approach based on generalized method of moments, which is efficient for the policy parameters. We propose a particular method based on recent developments on solving moment problems using neural networks and demonstrate the efficiency and regret benefits of this method empirically.
APA
Bennett, A. & Kallus, N.. (2020). Efficient Policy Learning from Surrogate-Loss Classification Reductions. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:788-798 Available from https://proceedings.mlr.press/v119/bennett20a.html.

Related Material