Off-policy Evaluation Beyond Overlap: Sharp Partial Identification Under Smoothness

Authors: Samir Khan, Martin Saveski, Johan Ugander

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness our methods on two semi-synthetic examples, and obtain informative and valid bounds that are tighter than those possible without smoothness assumptions.
Researcher Affiliation Academia 1Department of Statistics, Stanford University 2Information School, University of Washington 3Department of Management Science and Engineering, Stanford University.
Pseudocode No The paper does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code for replicating our results is available at https://github.com/skhan1998/lipschitz-ope.
Open Datasets Yes We use the Yeast data set from the UCI repository (Dua and Graff, 2017)... Our second experiment uses the Yahoo Webscope s featured news dataset, a standard benchmark for OPE algorithms (Yahoo!, 2011; Li et al., 2010; 2011).
Dataset Splits No The paper uses the Yeast and Yahoo datasets, but it does not specify explicit train/validation/test splits, percentages, or absolute sample counts for data partitioning to reproduce experiments. It mentions sampling and subsampling but not formal splits.
Hardware Specification No The paper does not provide any specific hardware details such as CPU/GPU models, processor types, or memory used for running the experiments.
Software Dependencies No The paper mentions fitting models using 'regularized logistic regression' and refers to 'kernel methods', but it does not specify any software names with version numbers (e.g., Python, PyTorch, scikit-learn, or specific solvers).
Experiment Setup No The paper describes general experimental settings, such as using a 'regularized logistic regression' for fitting models and varying parameters like 'L' and 'T' for the analysis of bounds. However, it does not provide concrete hyperparameter values (e.g., learning rate, batch size, regularization strength, optimizer settings) or detailed system-level training configurations for these models.