Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Non-stochastic Budgeted Online Pricing with Semi-Bandit Feedback

Authors: Xiang Liu, Hau Chan, Minming Li, Weiwei Wu, Long Tran-Thanh

AAAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct simulation experiments to show that the proposed policy outperforms the baseline algorithms. We empirically evaluate the proposed main policy GAP by comparing it with four state-of-the-art baseline algorithms from the literature.
Researcher Affiliation Academia 1Southeast University 2The Chinese University of Hong Kong 3University of Nebraska-Lincoln 4City University of Hong Kong 5The University of Warwick EMAIL, EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: The Granularity-based Pricing (GAP) Policy
Open Source Code No No explicit statement or link to open-source code for the methodology described in this paper is provided.
Open Datasets No We follow the standard setup for evaluating nonstochastic bandits (Zimmert, Luo, and Wei 2019; Alipour Fanid, Dabaghchian, and Zeng 2021). In particular, in our experiments the degree of the non-stochasticity can be quantified by the stochastically constrained adversaries. That is, the mean cost/value of each seller switches while staying unchanged for phases that are increasing exponentially in length (which is a common metric in non-stochastic bandit literature). The cost of seller si is uniformly set as follows, ci t [0.1, 0.2] if t belongs to an odd phase, [0.2, 1] otherwise and the value of each seller similarly switches between range [0.1, 0.2] and [0.2, 1].
Dataset Splits No The paper uses a simulated environment for experiments, where seller costs and values are defined programmatically. Therefore, traditional training/test/validation dataset splits are not applicable or mentioned.
Hardware Specification No The paper does not provide specific details about the hardware used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers needed to replicate the experiment.
Experiment Setup Yes By setting ϵ = c3/2 min/ B, η = ϵ/vmax and γ = ϵ/cmin, the expected α-regret of GAP is at most O n vmax B/cmin ln B where α = vmin (2 cmin)vmax . To evaluate the impact of the budget, we vary the budget in the range [5000, 40000] with the increment of 5000 and let B = 20000 by default. We also let the number of sellers, i.e., n, be selected from {5, 15, 25, 35, 45}, and let n = 25 by default.