On Strength Adjustment for MCTS-Based Programs
Authors: I-Chen Wu, Ti-Rong Wu, An-Jen Liu, Hung Guei, Tinghan Wei1222-1229
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We apply the above approach to the Go program ELF and present the experiment results. All the experiments are performed on machines equipped with one GTX 1080Ti GPU, one Intel Xeon E5-2683 v3 (14 cores in total), 2.6 GHz, 128 GB memory, and with Linux. All games are played with one second per move, using one GPU and six CPU cores. For each benchmark, 250 games are played against a baseline, ELF with and . [...] Table 1 shows the win rates and the relative Elo rating of the ELF versions with and with different against the baseline. Figure 1 shows the correlation between and the Elo ratings. [...] The above empirical results show that the strengths are highly correlated to . |
| Researcher Affiliation | Academia | Department of Computer Science, National Chiao Tung University, 1001 University Road, Hsinchu, Taiwan, ROC |
| Pseudocode | No | The paper describes methods and formulas but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions applying the method to the open source Go program 'ELF Open Go' and cites its GitHub repository (Tian et al. 2018; https://github.com/pytorch/ELF). However, this refers to a third-party project used by the authors, not the authors' own implementation code for their proposed strength adjustment method. |
| Open Datasets | No | The paper states it applies its approach to the 'Go program ELF' and mentions it 'follows the process of training Alpha Go Zero'. However, it does not provide concrete access information (link, DOI, specific repository, or formal citation for the dataset itself) for any specific dataset used for training or evaluating the proposed method. |
| Dataset Splits | No | No explicit training, validation, or test dataset splits are mentioned in the paper for the experiments conducted. The paper describes playing games to evaluate the strength adjustment method, e.g., '250 games are played against a baseline' and '100 games are played against each of the five opponents'. |
| Hardware Specification | Yes | All the experiments are performed on machines equipped with one GTX 1080Ti GPU, one Intel Xeon E5-2683 v3 (14 cores in total), 2.6 GHz, 128 GB memory, and with Linux. |
| Software Dependencies | No | The paper mentions 'Linux' as the operating system and 'ELF Open Go' as the game program, but does not provide specific version numbers for other software dependencies or libraries used for the experiments. |
| Experiment Setup | Yes | All games are played with one second per move, using one GPU and six CPU cores. For each benchmark, 250 games are played against a baseline, ELF with and . [...] for simplicity of analysis, 0.1 will be used as the threshold ratio, unless otherwise stated. [...] In our experiments for the method, is , approximately equivalent to 100 in Elo rating based on the linear regression in Figure 1, then decreased by a factor of for each game, with , equivalent to 8 in Elo rating. |