Computing Team-Maxmin Equilibria in Zero-Sum Multiplayer Extensive-Form Games

Authors: Youzhi Zhang, Bo An2318-2325

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm (IARAMDT) through experiments. We use CPLEX (version 12.9) to solve the linear program. All the algorithms are performed on a machine with 6-core 3.2GHz CPU and 16GB memory. Because the default gap for MILP in CPLEX is 10 6, we set ϵ1 = 10 6 with ϵ2 = 5 10 7 unless otherwise specified. We use BARON (version 19.3.24) (Khajavirad and Sahinidis 2018) to solve Problem (2) directly as a baseline. Other baselines include: 1) IARAMDT-1: at each iteration, we only select bilinear terms with the largest difference between wi(σTi) and ri(σi)wi+1(σTi+1) similar to the previous approach ( ˇCerm ak et al. 2018); 2) IARAMDT+1: we increase the power with 1 instead of Z at lines 10 and 11 of Algorithm 2, which is similar to many approaches ( ˇCerm ak et al. 2018; Andrade et al. 2019); 3) IRAMDT: we use RAMDT instead of ARAMDT; 4) IARMDT: we use MDT to replace AMDT; and 5) IRMDT: we increase the powers with 1 for all bilinear terms starting with Z = 1, which is the original MDT approach (Kolodziej, Castro, and Grossmann 2013; Wang, Guo, and An 2017). Our experiments run on the standard games, the multiplayer Kuhn poker and the Leduc Hold em poker games (see Farina et al. (2018) for their rules), where n Kr and n Lr denote an n-player Kuhn instance with r ranks (i.e., r cards) and an n-player Leduc instance with r ranks (i.e., 2r cards),
Researcher Affiliation Academia Youzhi Zhang, Bo An School of Computer Science and Engineering, Nanyang Technological University, Singapore {yzhang137, boan}@ntu.edu.sg
Pseudocode Yes Algorithm 1: Generate associated constraints [...] Algorithm 2: Iterative ARAMDT (IARAMDT)
Open Source Code No The paper does not provide any explicit statement about making the source code available or include a link to a code repository.
Open Datasets Yes Our experiments run on the standard games, the multiplayer Kuhn poker and the Leduc Hold em poker games (see Farina et al. (2018) for their rules), where n Kr and n Lr denote an n-player Kuhn instance with r ranks (i.e., r cards) and an n-player Leduc instance with r ranks (i.e., 2r cards), respectively.
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits with percentages, sample counts, or references to predefined splits for the game environments used in the experiments.
Hardware Specification Yes All the algorithms are performed on a machine with 6-core 3.2GHz CPU and 16GB memory.
Software Dependencies Yes We use CPLEX (version 12.9) to solve the linear program. [...] We use BARON (version 19.3.24) (Khajavirad and Sahinidis 2018) to solve Problem (2) directly as a baseline.
Experiment Setup Yes Because the default gap for MILP in CPLEX is 10 6, we set ϵ1 = 10 6 with ϵ2 = 5 10 7 unless otherwise specified.