Transfer Learning for Nonparametric Contextual Dynamic Pricing
Authors: Fan Wang, Feiyu Jiang, Zifeng Zhao, Yi Yu
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive numerical experiments validate our approach, demonstrating its superiority over existing methods and highlighting its practical utility in real-world applications. In this section, we conduct numerical experiments to support our theoretical findings. Synthetic and real data analysis are in Sections 5.1 and 5.2, respectively. |
| Researcher Affiliation | Academia | 1Department of Statistics, University of Warwick 2School of Management, Fudan University 3Mendoza College of Business, University of Notre Dame. Correspondence to: Fan Wang <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Transfer Learning for Dynamic Pricing (TLDP) |
| Open Source Code | Yes | The code and datasets are available online1. 1https://github.com/chrisfanwang/dynamic-pricing |
| Open Datasets | Yes | The code and datasets are available online1. 1https://github.com/chrisfanwang/dynamic-pricing ... using the auto loan dataset (Phillips et al., 2015) |
| Dataset Splits | Yes | For this study, the East South Central division (8,062 applications) is designated as the target domain due to its smallest data size. ... The results are averaged over 100 simulations, with each simulation randomly selecting 90% of the target data as the test data. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud computing instances used for running the experiments. It only mentions 'simulation studies' and 'numerical experiments'. |
| Software Dependencies | No | The true reward function is approximated using a random forest model implemented in R (R Core Team, 2021), leveraging the randomForest package (Liaw & Wiener, 2002), trained on the target data. While R and the randomForest package are mentioned with citations, specific version numbers for these software components are not provided. |
| Experiment Setup | Yes | Specifically, we set CI = 1 and compute r from (13) using the true values of the exploration coefficient κ, the transfer exponent γ and Cr = 1/4. ... For the ABE algorithm, we set M = 0.1, the constant used to define the maximal number observed in a level-k bin in the partition, as suggested by Chen & Gallego (2020). ... For Ex UCB, parameter choices include the phase exponent β = 2/3, the UCB exponent γ = 1/6, the exploration phase constant C1 = 1, the discretization constant C2 = 20 and the regularization parameter in UCB λ = 0.1, as suggested by Luo et al. (2022). In addition, we fix the maximum price pmax = 1 and the maximum possible revenue B = 1 for the Ex UCB implementation. |