Reduced Space and Faster Convergence in Imperfect-Information Games via Pruning
Authors: Noam Brown, Tuomas Sandholm
ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that BRP results in a factor of 7 reduction in space, and the reduction factor increases with game size. |
| Researcher Affiliation | Academia | 1Computer Science Department, Carnegie Mellon University, Pittsburgh, PA, USA. Correspondence to: Noam Brown <noamb@cs.cmu.edu>, Tuomas Sandholm <sandholm@cs.cmu.edu>. |
| Pseudocode | No | No pseudocode or algorithm blocks are explicitly provided in the paper. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | The experiments are conducted on Leduc Hold em (Southey et al., 2005) and Leduc-5 (Brown & Sandholm, 2015a). |
| Dataset Splits | No | The paper evaluates performance within game environments (Leduc Hold'em, Leduc-5) and discusses convergence over iterations, but does not define traditional train/validation/test dataset splits as it's not a typical supervised learning setup with fixed datasets. |
| Hardware Specification | No | The paper uses 'Nodes touched' as a hardware and implementation-independent proxy for time and does not provide specific details on the hardware used for experiments. |
| Software Dependencies | No | The paper discusses various algorithms (CFR, CFR+, RM, Fictitious Play) but does not specify any software dependencies or library version numbers used for implementation. |
| Experiment Setup | Yes | Figure 1 and Figure 2 show the reduction in space needed to store the average strategy and regrets for BRP for various values of the constant threshold C, where an action s probability is set to zero if it is reached with probability less than C T in the average strategy, as we explained in Section 3.1. In both games, a threshold between 0.01 and 0.1 performed well in both space and number of iterations, with the lower thresholds converging somewhat faster and the higher thresholds reducing space faster. |