Fast Payoff Matrix Sparsification Techniques for Structured Extensive-Form Games

Authors: Gabriele Farina, Tuomas Sandholm4999-5007

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally compare the sparsification techniques introduced in Section 4 on eight River endgames that were actually played in the Brains vs AI competition where superhuman performance was reached by an AI, Libratus, against four top specialist professional players in no-limit Texas hold em in January 2017. Full results are in available in Table 1.
Researcher Affiliation Collaboration Gabriele Farina1, Tuomas Sandholm1,2,3,4 1 Department of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213 2 Strategy Robot, Inc. 3 Optimized Markets, Inc. 4 Strategic Machine, Inc. {gfarina,sandholm}@cs.cmu.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include an unambiguous statement or link for the release of the source code for the described methodology.
Open Datasets No The paper mentions using "eight River endgames that were actually played in the Brains vs AI competition" and that "Noam Brown for providing the poker endgames". However, it does not provide concrete public access information (link, DOI, formal citation) to this dataset.
Dataset Splits No The paper does not provide specific dataset split information (percentages, sample counts, or detailed splitting methodology) for training, validation, or testing.
Hardware Specification Yes All experiments were conducted on a computer with 32GB of RAM and an Intel CPU with 16 (virtual) cores, each with a nominal speed of 2.40GHz. Our GPU version of the algorithm was implemented within Nvidia s CUDA framework and run on a laptop-grade Quadro T2000 GPU.
Software Dependencies No The paper mentions using "C++", "Eigen library", "Gurobi", and "Cusparse libraries" but does not provide specific version numbers for any of these software dependencies.
Experiment Setup Yes We ran the iterative algorithm of Zhang and Sandholm (2020) with a fixed random seed and a cap on the number of sparsifying iterations set to 1000. We investigate computing least-exploitable deterministic strategies in all eight river endgames, using the full Libratus betting abstraction in the three smallest games and endgame 4, and the smaller betting abstraction used by Zhang and Sandholm (2020) in the remaining games as Gurobi struggled to solve the larger games with the larger abstraction.