Batched Thompson Sampling

Authors: Cem Kalkanli, Ayfer Ozgur

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experimental Setup, 6 Experiments and Results, We evaluate BT-OFUL and BT-UCB against the batch Thompson Sampling (BTS) algorithm by considering both synthetic datasets and real-world datasets.
Researcher Affiliation Academia Yuehan Liu, Dongdong Ge, Weiran Shen Peking University
Pseudocode Yes Figure 1: Algorithm 1: BT-TS with fixed batch size, Figure 2: Algorithm 2: BT-OFUL, Figure 3: Algorithm 3: BT-UCB
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of their methodology.
Open Datasets Yes We evaluate BT-OFUL and BT-UCB against the batch Thompson Sampling (BTS) algorithm by considering both synthetic datasets and real-world datasets. For real-world datasets, we use the Yahoo! Front Page Today Module (FP.Today) dataset [24] and MovieLens 1M dataset.
Dataset Splits No For the Yahoo! Today Front Page dataset, we use the first 200,000 samples for training and the next 100,000 samples for testing. No explicit mention of a 'validation' split.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) are mentioned for running the experiments.
Software Dependencies No No specific software dependencies with version numbers are mentioned in the paper.
Experiment Setup Yes For the synthetic datasets, we use a fixed learning rate of 0.01 for all algorithms. For the Yahoo! Today Front Page dataset, we set the batch size K = 1000, λ = 1, and η = 0.01. Similar parameters are used for MovieLens 1M dataset: batch size K=1000, λ = 1, η = 0.01.