Doubly Robust Counterfactual Classification

Authors: Kwangho Kim, Edward Kennedy, Jose Zubizarreta

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We study the empirical performance of our methods by simulation and apply them for recidivism risk prediction.
Researcher Affiliation Academia Kwangho Kim Harvard Medical School kkim@hcp.med.harvard.edu Edward H. Kennedy Carnegie Mellon University edward@stat.cmu.edu José R. Zubizarreta Harvard University zubizarreta@hcp.med.harvard.edu
Pseudocode Yes Algorithm 1: Doubly robust estimator for counterfactual classification
Open Source Code No Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [No]
Open Datasets Yes Next we apply our method for recidivism risk prediction using the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) dataset 2. https://github.com/propublica/compas-analysis
Dataset Splits Yes We use sample splitting as described in Algorithm 1 with K = 2 splits.
Hardware Specification No Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [N/A]
Software Dependencies No The paper mentions 'SUPERLEARNER R package' and 'NLOPTR R package' as well as 'Sto Go' and 'BOBYQA' algorithms, but it does not specify version numbers for these software components.
Experiment Setup Yes For nuisance estimation we use the cross-validation-based Super Learner ensemble via the SUPERLEARNER R package to combine generalized additive models, multivariate adaptive regression splines, and random forests. We use sample splitting as described in Algorithm 1 with K = 2 splits... To solve b P, we first use the Sto Go algorithm [40] via the NLOPTR R package as it has shown the best performance in terms of accuracy in the survey study of [35]. After running the Sto Go, we then use the global optimum as a starting point for the BOBYQA local optimization algorithm [41] to further polish the optimum to a greater accuracy. We use sample sizes n = 1k, 2.5k, 5k, 7.5k, 10k and repeat the simulation 100 times for each n.