Structured Dynamic Pricing: Optimal Regret in a Global Shrinkage Model
Authors: Rashmi Ranjan Bhuyan, Adel Javanmard, Sungchul Kim, Gourab Mukherjee, Ryan A. Rossi, Tong Yu, Handong Zhao
JMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct simulation experiments across a wide range of regimes as well as real-world networks based studies and report encouraging performance for our proposed method. |
| Researcher Affiliation | Collaboration | Rashmi Ranjan Bhuyan EMAIL Department of Data Sciences and Operations, Marshall School of Business University of Southern California, Los Angeles, CA 90089 , USA Adel Javanmard EMAIL Department of Data Sciences and Operations, Marshall School of Business University of Southern California, Los Angeles, CA 90089 , USA Sungchul Kim EMAIL Adobe Research, 345 Park Avenue, San Jose, CA 95110 USA Gourab Mukherjee EMAIL Department of Data Sciences and Operations, Marshall School of Business University of Southern California, Los Angeles, CA 90089 , USA Ryan A. Rossi EMAIL Adobe Research, 345 Park Avenue, San Jose, CA 95110 USA Tong Yu EMAIL Adobe Research, 345 Park Avenue, San Jose, CA 95110 USA Handong Zhao EMAIL Adobe Research, 345 Park Avenue, San Jose, CA 95110 USA |
| Pseudocode | Yes | Algorithm 1 PSGD based Dynamic Pricing Policy |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It only mentions licensing and attribution for the paper itself. |
| Open Datasets | Yes | Here, we consider L = 48 segments. Each segments constitute a US state. For ease of analysis and presentation, we remove Hawaii and Alaska from the analysis. We choose 15 demographic and socio-economic variables such as percentage of residents in the age group 5-65, average income, unemployement rates, etc for making the network matrix W among the L segments. We use an RBF kernel of width two and threshold the resultant network at 0.05 level, i.e., edges with weight less than 0.05 are deleted from the network. Figure 4 shows the network. We generate the covariates xlt using standard exponential distribution." and "Bureau (2008). Statistical Abstract of the United States 2008." |
| Dataset Splits | No | The paper describes various experimental setups and data generation processes (e.g., 'imbalanced design' in Set-up 4, 'sampling heterogeneity') but does not specify how data is split into training, test, or validation sets for the main regret analysis of the pricing policy. A 'train/test split' is mentioned only in the context of cross-validation for determining a bandwidth for the interaction matrix W, not for evaluating the model's performance. |
| Hardware Specification | No | The paper does not explicitly describe the hardware specifications (e.g., CPU, GPU models, memory) used for running its experiments or simulations. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | Set-up 1. Consider L = 10 segments and a time-invariant sampling policy with different sampling rates across two segment groups: for any t ≥ 1, nlt = 50 for l = 1, . . . , 5 and nlt = 200 for l = 6, . . . , 10. We use a time-invariant network W that was generated using radial basis function (RBF) kernel of width one on independent standard Gaussian feature vectors, drawn from input space R10. We use bivariate covariates xlt generated from standard exponential distribution and set τ = 1, σ = 1 in (2)-(6). The price sensitivity and the customer preferences are assumed as β1 = 0.4 and µ1 = (0.1, 0.15). With change in time the parameters change as follows, βt+1 = βt + δtβ; µt+1 = µt + δtµ, where δtβ = t−bZt/(10|Zt|) and δtµ = t−bZt/(10 Zt ) where Zt and Zt are standard Gaussian random variables of dimension 1 and 2 respectively. We set ρt to 0.5 for all t ≥ 1 and consider three cases, b = 0.5, 1, ∞, for the temporal variations across βt and µt. |