Solving Imperfect-Information Games via Discounted Regret Minimization

Authors: Noam Brown, Tuomas Sandholm1829-1836

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments are run for 32,768 iterations for HUNL subgames and 8,192 iterations for Goofspiel. We measure the average exploitability of the two players. Our experiments show that LCFR can dramatically improve performance over CFR+ over reasonable time horizons in certain games.
Researcher Affiliation Academia Noam Brown Computer Science Department Carnegie Mellon University noamb@cs.cmu.edu Tuomas Sandholm Computer Science Department Carnegie Mellon University sandholm@cs.cmu.edu
Pseudocode No The paper describes algorithms and their modifications using mathematical equations and textual descriptions but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper refers to an external GitHub link for "Libratus Endgames" (https://github.com/CMU-EM/Libratus Endgames) which pertains to the subgames used as input, but it does not provide source code for the novel CFR variants presented in the paper itself.
Open Datasets Yes We conduct our HUNL experiments on four subgames generated by Libratus 1. The subgames were selected prior to testing. 1https://github.com/CMU-EM/Libratus Endgames In addition to HUNL subgames, we also consider a version of the game of Goofspiel (limited to just five cards per player).
Dataset Splits No The paper describes running simulations for a fixed number of iterations and measuring exploitability, but it does not define traditional machine learning dataset splits (e.g., training, validation, test sets) for reproduction.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instances) used to conduct the experiments.
Software Dependencies No The paper does not specify any software dependencies, libraries, or solvers with their version numbers that are needed to replicate the experiments.
Experiment Setup Yes Our experiments are run for 32,768 iterations for HUNL subgames and 8,192 iterations for Goofspiel. However, we found that setting α = 3/2, β = 0, and γ = 2 led to performance that was consistently stronger than CFR+. Thus, when we refer to DCFR with no parameters listed, we assume this set of parameters are used.