Optimal Comparator Adaptive Online Learning with Switching Cost
Authors: Zhiyu Zhang, Ashok Cutkosky, Yannis Paschalidis
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Concluding these theoretical results, our OLO algorithm is applied to a portfolio management task with transaction costs (Appendix D). Numerical results support its superiority over the existing approach [ZCP22a]. |
| Researcher Affiliation | Academia | Zhiyu Zhang Boston University zhiyuz@bu.edu Ashok Cutkosky Boston University ashok@cutkosky.com Ioannis Ch. Paschalidis Boston University yannisp@bu.edu |
| Pseudocode | Yes | We will specifically consider Algorithm 1, which is a variant based on the discrete derivative SV , cf. (1). Algorithm 1 One-dimensional unconstrained OLO with switching costs. Most notably, we present an algorithm (Algorithm 5) for 1D bounded domain: if the domain has diameter D, then the switching cost alone of this algorithm is bounded by O(D τ) on any time interval of length τ. |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] In supplemental material. |
| Open Datasets | Yes | The historical price data of assets from the S&P 500 index are downloaded from Yahoo Finance. |
| Dataset Splits | No | The paper specifies a temporal split for initialization and sequential predictions ('The first 50 days are used for initialization, and the remaining days are used for sequential predictions.') but does not explicitly define distinct training, validation, and test sets with percentages or sample counts for traditional model validation. |
| Hardware Specification | No | Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No] Our experiments are not computationally demanding. |
| Software Dependencies | No | The paper does not specify software dependencies with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9'). |
| Experiment Setup | Yes | Our experiments are based on the same setup as [LWZ18] and [MR22]. We set the initial wealth W0 = 100. We also set the hyperparameter C = 0.01 and the Lipschitz constant G = 1 as in [ZCP22a]. |