Solving Large Sequential Games with the Excessive Gap Technique
Authors: Christian Kroer, Gabriele Farina, Tuomas Sandholm
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show experimental results on our variant of the excessive gap technique as well as a prior version. We now present experimental results on running all the previously described algorithms on a GPU. |
| Researcher Affiliation | Academia | Christian Kroer, Gabriele Farina, and Tuomas Sandholm Department of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 {ckroer,gfarina,sandholm}@cs.cmu.edu |
| Pseudocode | Yes | Algorithm 1 EGT(DGF-center xω, DGF weights µx, µy, and ϵ > 0) Algorithm 2 STEP(µx, µy, x, y, τ) Algorithm 3 EGT/AS(DGF-center xω, DGF weights µx, µy, and ϵ > 0) Algorithm 4 DECR(µx, µy, x, y, τ) |
| Open Source Code | No | The paper states 'All code was implemented in C++ using CUDA for GPU operations, and cu SPARSE for the sparse payoff matrix.', but does not provide a specific link or explicit statement about the release of their own implementation's source code. |
| Open Datasets | No | The paper states 'Our experiments are conducted on real large-scale river endgames faced by the Libratus AI [4].' and refers to 'endgames extracted from Libratus play', but does not provide concrete access information (link, DOI, specific repository name, or formal citation with authors/year) for a publicly available dataset. |
| Dataset Splits | No | The paper describes experiments on 'Endgame 2' and 'Endgame 7' but does not specify any training, validation, or test dataset splits; these are described as specific game instances rather than data partitions. |
| Hardware Specification | Yes | All experiments were run on a Google Cloud instance with an NVIDIA Tesla K80 GPU with 12GB available. |
| Software Dependencies | No | All code was implemented in C++ using CUDA for GPU operations, and cu SPARSE for the sparse payoff matrix. This lists software names but does not provide specific version numbers. |
| Experiment Setup | Yes | A subgame is structured and parameterized as follows. The game is parameterized by the conditional distribution over hands for each player, current pot size, board state (5 cards dealt to the board), and a betting abstraction. Then, Libratus has the choice of folding, checking, or betting by a number of multipliers of the pot size: 0.25x, 0.5x, 1x, 2x, 4x, 8x, and all-in. |