Recursive Partitioning for Personalization using Observational Data

By Nathan Kallus. 2016

102 citations recorded by the source corpus.

We study the problem of learning to choose from m discrete treatment options (e.g., news item or medical drug) the one with best causal effect for a particular instance (e.g., user or patient) where the training data consists of passive observations of covariates, treatment, and the outcome of the treatment. The standard approach to this problem is regress and compare: split the training data by treatment, fit a regression model in each split, and, for a new instance, predict all m outcomes and pick the best. By reformulating the problem as a single learning task rather than m separate ones, we propose a new approach based on recursively partitioning the data into regimes where different treatments are optimal. We extend this approach to an optimal partitioning approach that finds a globally optimal partition, achieving a compact, interpretable, and impactful personalization model. We develop new tools for validating and evaluating personalization models on observational data and use these to demonstrate the power of our novel approaches in a personalized medicine and a job training application.

Subfields: Artificial Intelligence, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Balanced Policy Evaluation and Learning (2017, methodology): “5), the direct method with Gaussian process regression gradient boosted trees (both sklearn defaults), weighted SVM classification using IPW and DR weights (details in supplement), SNPOEM [43], PF [23], and our balanced policy learner (1) with parameters as in Ex.”
Generalized random forests (2016, background): “…for other problems, including Athey and Imbens (2016) and Su et al. (2009) for treatment effect estimation, Beygelzimer and Langford (2009) and Kallus (2016) for personalized policy allocation, and Gordon and Olshen (1985), LeBlanc and Crowley (1992), Molinaro, Dudoit and Van der Laan (2004)…”
Policy Evaluation and Optimization with Continuous Treatments (2018, background): “Previous work on predicting dosage policies [2, 11] has evaluated accuracy based on prediction of the correct category of dosage, “low” (<21 mg/wk), “medium” (> 21 mg/wk,< 49 mg/wk) or “high” (> 49 mg/wk).”
Efficient Policy Learning (2017, methodology): “, inverse propensity weighting as used in Kitagawa and Tetenov (2015) or Swaminathan and Joachims (2015), such a slicing argument may not work.”
Residual Unfairness in Fair Machine Learning from Prejudiced Data (2018, background): “We sometimes call Z = 1 the logging policy in analogy to logged-bandit learning (Kallus, 2017; Swaminathan & Joachims, 2015), where the implementation of a (often unknown) historical policy resulted in limited bandit feedback on outcomes.”
Confounding-Robust Policy Improvement (2018, methodology): “This leads to strong generalization guarantees (Kallus 2017b, Kitagawa and Tetenov 2015, Wager and Athey 2017a) and can also allow one to incorporate domain-specific constraints that favor simple prescriptive decision policies that are interpretable, implementable, and/or satisfy operational…”
Optimal trees for prediction and prescription (2018, methodology): “Continuing in the spirit of adapting machine learning approaches, Kallus [70] proposes personalization trees (and forests), which adapt regular classification trees [25] to directly optimize the prescription error.”
Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks (2020, methodology): “, 2019), building sub-populations using tree based methods (Chipman et al., 2010; Athey and Imbens, 2016; Wager and Athey, 2018; Kallus, 2017) or building balancing representations between patients receiving the different treatments (Johansson et al.”
Market Segmentation Trees (2019, methodology): “Kallus (2017) and Bertsimas et al. (2019) propose methodology for training decision trees for segmenting customers and personalizing treatments across the resulting segments.”
Optimization of Tree Ensembles (2017, background): “In pricing, data on historical prices and demand observed at those prices is often used to build a predictive model of demand as a function of price and to then determine the optimal price (see for example Besbes et al. 2010, Bertsimas and Kallus 2016).”
Decision Trees for Decision-Making under the Predict-then-Optimize Framework (2020, unknown): “Kallus (2017) uses a loss function for training their trees which maximizes the eﬃcacy of the recommended treatments rather than minimizing prediction error.”
Optimal Prescriptive Trees (2019, methodology): “(6) We note that, when µ 1, we obtain the same objective function as Kallus (2017), whichmeans that this objective is an unbiased and consistent estimator for the prescription error.”
Policy Evaluation with Latent Confounders via Optimal Balance (2019, background): “In addition some other approaches include recursive partitioning [14].”
Bayesian Nonparametric Causal Inference: Information Rates and Learning Algorithms (2017, unknown): “There is a growing interest in using machine learning methods to infer the individualized causal effects of medical treatments; this interest manifests in recent initiatives such as STRATOS [4], which focuses on guiding observational medical research, in addition to various recent works on causal ef”