MCTS-Minimax Hybrids with State Evaluations
Authors: Hendrik Baier, Mark H. M. Winands
JAIR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Results show that the use of enhanced minimax for computing node priors results in the strongest MCTS-minimax hybrid investigated in the three test domains of Othello, Breakthrough, and Catch the Lion. This hybrid, called MCTS-IP-M-k, also outperforms enhanced minimax as a standalone player in Breakthrough, demonstrating that at least in this domain, MCTS and minimax can be combined to an algorithm stronger than its parts. |
| Researcher Affiliation | Academia | Hendrik Baier EMAIL Digital Creativity Labs, University of York, York, UK; Mark H. M. Winands EMAIL Games and AI Group, Department of Data Science and Knowledge Engineering, Maastricht University, Maastricht, The Netherlands |
| Pseudocode | No | The paper describes the MCTS algorithm and its variants (MCTS-IR, MCTS-IC, MCTS-IP) through textual descriptions and illustrative figures (Figure 1, 2, 3, 4), but it does not present any explicit pseudocode blocks or algorithms labeled as such. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide links to any code repositories. |
| Open Datasets | No | The paper uses well-known games (Othello, Catch the Lion, Breakthrough) as test domains and describes the evaluation functions used for these games. However, it does not provide concrete access information (e.g., specific links, DOIs, repository names, or formal citations) for any specific dataset of game states or plays used in the experiments. |
| Dataset Splits | No | The paper describes experiments where algorithms played a certain number of games against baselines (e.g., "Each condition played 1000 games per domain against the baseline player." and "In 2000 additional games against the baseline..."). This indicates evaluation methodology but does not provide specific training/test/validation dataset splits as would typically be found in machine learning experiments with pre-existing datasets. |
| Hardware Specification | Yes | Computation time was 1 second per move. In order to make a comparison of the relative computational costs of all algorithms possible, we are also listing their speed in terms of rollouts per second (rps), measured on an Intel Core i7-3667U CPU at 2.0 GHz using the average over 10 1-second searches from the initial board position in each game. |
| Software Dependencies | No | The paper mentions specific algorithms and techniques like "MCTS-Solver extension", "UCB1-TUNED policy", and "αβ pruning". However, it does not specify version numbers for any software libraries, frameworks, or tools used for implementation. |
| Experiment Setup | Yes | Optimal MCTS parameters such as the exploration factor C were determined once for MCTS-Solver in each game and then kept constant for both MCTS-Solver and the MCTS-minimax hybrids during testing. C was set to 0.7 in Othello and Catch the Lion, and 0.8 in Breakthrough. ... MCTS-IR-E was tested for ϵ {0, 0.05, 0.1, 0.2, 0.5}. ... MCTS-IR-M was tested for d {1, . . . , 4}... MCTS-IC-E was tested for m {0, . . . , 5}. ... MCTS-IP-E was tested for all combinations of n {0, 1, 2} and γ {50, 100, 250, 500, 1000, 2500, 5000}. |