Approximate Dynamic Programming for Two-Player Zero-Sum Markov Games

Authors: Julien Perolat, Bruno Scherrer, Bilal Piot, Olivier Pietquin

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we demonstrate experimentally the performance of AGPIQ on a simultaneous two-player game, namely Alesia.In this section, AGPI-Q is tested on the Alesia game described in Sec. 2.5 where we assume that both players start with a budget n = 20. As a baseline, we use the exact solution of the problem provided by VI. We have run the algorithm for K = 10 iterations and for m {1, 2, 3, 4, 5} evaluation steps. We have considered different sample set sizes, N = 2500, 5000, 10000. Each experiment is repeated 20 times.
Researcher Affiliation Academia (1)Univ. Lille, CRISt AL, Seque L team, France (2)Inria, Villers-l es-Nancy, F-54600, France (3)Institut Universitaire de France (IUF), France
Pseudocode Yes Algorithm 1 AGPI Q for Batch sample Input: ((xj, aj 1, aj 2), rj, x j)j=1,...,N some samples, q0 = 0 a Q-function, F an hypothesis space for k=1,2,...,K do Greedy step: for all j do aj = arg max a min b qk 1(x j, a, b) (solving a matrix game) end for Evaluation step: qk,0 = qk 1 for i=1,...,m do for all j do qj = r(xj, aj 1, aj 2) + γ minb qk,i 1(x j, aj, b) end for qk,i = arg minq F j=1 l(q(xj, aj 1, aj 2), qj) Where l is a loss function. qk = qk,m end for end for output q K
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper.
Open Datasets No The paper describes the Alesia game and how samples were generated, but does not provide concrete access information (link, DOI, repository name, or formal citation with authors/year) for a publicly available or open dataset used in the experiments.
Dataset Splits No The paper describes generating N uniform samples and using them in a batch setting, but does not specify explicit training, validation, or test dataset splits for these samples.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions using CART trees and linear programming techniques, but does not specify any software names with version numbers for replication.
Experiment Setup Yes We have run the algorithm for K = 10 iterations and for m {1, 2, 3, 4, 5} evaluation steps. We have considered different sample set sizes, N = 2500, 5000, 10000. Each experiment is repeated 20 times.