A Fine-grained Analysis of Fitted Q-evaluation: Beyond Parametric Models

Authors: Jiayi Wang, Zhengling Qi, Raymond K. W. Wong

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct a simulation study to illustrate the behavior of the error |ˆν(π) ν(π)| with respect to n and T. The goal here is to provide empirical evidence of our theoretical results, and so we use a relatively simple simulation setup for the purpose of clear demonstration.
Researcher Affiliation Academia 1Department of Mathematical Sciences, University of Texas at Dallas, Richardson, USA 2School of Business, The George Washington University, Washington, D.C., USA 3Department of Statistics, Texas A&M University, College Station, USA.
Pseudocode No The paper describes the FQE method and provides mathematical formulations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statements about making its source code publicly available, nor does it provide a link to a code repository.
Open Datasets No The paper uses synthetic data generated via a specified model rather than a publicly available dataset. It describes: 'The state variable is a one-dimensional continuous variable and the action is a binary variable, i.e., At = {0, 1} for all t. The initial state follows the uniform distribution within [ 2, 2]. The transition dynamics are given by Si,t+1 = (2Ai,t 1)f(Si,t)...'
Dataset Splits No The paper describes the use of leave-one-out cross-validation to decide the number of basis functions (K) but does not specify any explicit training, validation, or test dataset splits for the simulated data.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, or memory) used to run the simulation studies.
Software Dependencies No The paper mentions using 'cubic B-spline' for constructing basis functions but does not specify any software packages or their version numbers used for implementation or analysis.
Experiment Setup Yes We conduct a simulation study to illustrate the behavior of the error |ˆν(π) ν(π)| with respect to n and T. ... We evaluate values with n = 200, 400, . . . , 2000, and T = 20, 40, . . . , 200. We use cubic B-spline to construct basis functions at every step t. The knots are placed at evenly distributed percentiles of samples. ... we fix K = 3n1/5. For the second approach, we use leave-one-out cross-validation to decide K at every step.