Stochastic Regret Minimization in Extensive-Form Games

Authors: Gabriele Farina, Christian Kroer, Tuomas Sandholm

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show extensive experiments on five games, where some variants of our methods outperform MCCFR.In this section we perform numerical simulations to investigate the practical performance of several stochastic regretminimization algorithms.
Researcher Affiliation Collaboration 1Computer Science Department, Carnegie Mellon University, Pittsburgh PA 15213 2IEOR Department, Columbia University, New York NY 10027 3Strategic Machine, Inc. 4Strategy Robot, Inc. 5Optimized Markets, Inc.
Pseudocode Yes Algorithm 1: Efficient implementation of the external sampling gradient estimator
Open Source Code No The paper does not contain any explicit statement about providing open-source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets No The paper describes the games used (Leduc poker, Goofspiel, Search, Battleship) but does not provide specific links, DOIs, repository names, or explicit citations that directly lead to publicly available dataset files for these games. It mentions they are 'standard parametric benchmark games' but lacks concrete access details for the data itself.
Dataset Splits No The paper describes game environments used for experiments but does not provide information about training/validation/test dataset splits, as the experiments are not based on traditional machine learning datasets that are partitioned this way. It refers to iterations of algorithms within game environments.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as programming languages, libraries, or solvers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes For each game, we try four choices of stepsize η in FTRL and OMD: 0.1, 1, 10, 100.For that gradient estimator we drew 100 outcome samples per gradient estimate, and use the empirical mean of those 100 samples as our estimate.All algorithms are run until the number of nodes touched corresponds to 50 full tree traversals (or, equivalently, 25 iterations of deterministic CFR or CFR+).