Uplifting Bandits

Authors: Yu-Guan Hsieh, Shiva Kasiviswanathan, Branislav Kveton

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and real-world datasets show the benefit of methods that estimate the uplifts over policies that do not use this structure. 8 Numerical Experiments
Researcher Affiliation Collaboration Yu-Guan Hsieh University of Grenoble Alpes yu-guan.hsieh@univ-grenoble-alpes.fr Shiva Prasad Kasiviswanathan Amazon kasivisw@gmail.com Branislav Kveton Amazon bkveton@amazon.com
Pseudocode Yes Algorithm 1 UPUCB; Algorithm 2 UPUCB-n Aff; Algorithm 3, Appendix A; Algorithm 4 in Appendix A
Open Source Code Yes 3. If you ran experiments... (a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] Provided in the supplmental material.
Open Datasets Yes We use the Criteo Uplift Prediction Dataset [13] with visit as the outcome variable to build a Bernoulli uplifting bandit... [13] Eustache Diemert, Artem Betlei, Christophe Renaudin, and Massih-Reza Amini. A large scale benchmark for uplift modeling. In International Conference on Knowledge Discovery and Data Mining. ACM, 2018.
Dataset Splits No The paper describes how samples are used to build the model ('sample 10^5 examples from the dataset, and use K-means to partition these samples'), but it does not specify explicit train/validation/test splits, percentages, or a cross-validation methodology.
Hardware Specification Yes All experiments are conducted using a standard desktop machine with Intel Core i7-2600 and 16GB of RAM.
Software Dependencies No The code is written in Python 3.9 and uses standard libraries such as numpy and scipy. However, specific version numbers for numpy and scipy are not provided, only the Python version.
Experiment Setup No The paper states that the algorithms are 'tuned...for the parameters that yield the best average performance,' but it does not provide the specific hyperparameter values or detailed training configurations (e.g., learning rates, batch sizes, number of epochs) used for the experiments.