Thompson Sampling for High-Dimensional Sparse Linear Contextual Bandits
Authors: Sunrit Chakraborty, Saptarshi Roy, Ambuj Tewari
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive simulations demonstrate the improved performance of our proposed algorithm over existing ones. 5. Numerical Experiments In both simulations and real data experiments, we present results corresponding to λt = 1 for all t [T]. 5.1. Synthetic data In this section, we illustrate the performance of the VBTS algorithm on a simulated data set. 5.2. Real data gravier Breast Carcinoma Data We consider breast cancer data gravier (microarray package in R) for 168 patients to predict metastasis of breast carcinoma based on 2905 gene expressions (bacterial artificial chromosome or BAC array). |
| Researcher Affiliation | Academia | 1Department of Statistics, University of Michigan, Ann Arbor, USA. |
| Pseudocode | Yes | Algorithm 1 Thompson Sampling Algorithm; Algorithm 2 Variational Bayes Thompson Sampling |
| Open Source Code | Yes | Codes are available online: Github link. |
| Open Datasets | Yes | We consider breast cancer data gravier (microarray package in R) for 168 patients to predict metastasis of breast carcinoma based on 2905 gene expressions (bacterial artificial chromosome or BAC array). (Gravier et al., 2010) |
| Dataset Splits | No | The paper does not provide explicit training, validation, or test dataset splits. For the real data experiment, it describes a setup where samples are drawn in each round for a contextual bandit problem: "In each round, the environment randomly draws one sample from each class and composes a set of contexts of 2 samples." This is not a fixed dataset split for reproduction purposes. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications, or cloud instance types) used for running the experiments. It only mentions "Table 2 shows the mean execution time..." without specifying the hardware on which these times were measured. |
| Software Dependencies | No | The paper mentions using the "sparsevb package (Clara et al., 2021) in R" and states that it uses the "Coordinate Ascent Variational Inference (CAVI) algorithm proposed in (Ray & Szab o, 2021)". However, it does not explicitly provide specific version numbers for R or the 'sparsevb' package within the main text or a dedicated software dependencies section. While the citation for 'sparsevb' includes a version number, this is not directly stated as a required dependency version in the paper's body. |
| Experiment Setup | Yes | In both simulations and real data experiments, we present results corresponding to λt = 1 for all t [T]. ...we also present the simulation results for synthetic data experiments with λt = λ t for λ {0.2, 0.3, 0.4, 0.5}... We set the number of arms K = 10 and we generate the context vectors {xi(t)}K i=1 from multivariate d-dimensional Gaussian distribution Nd(0, Σ), where Σij = ρ|i j| 1 and ρ = 0.3. We consider d = 1000 and the sparsity s = 5. |