Searching the whole tree while extracting features for each move is totally impractical given our branching factor and cost of feature extraction. However, searching for living groups is much cheaper. Electing to modify search to search terminal states to maxdepth, but only extract features and evaluate moves with the classifier for moves one level deep, assuming there are no terminal "good" states to choose.
Searching the whole tree while extracting features for each move is totally impractical given our branching factor and cost of feature extraction. However, searching for living groups is much cheaper. Electing to modify search to search terminal states to maxdepth, but only extract features and evaluate moves with the classifier for moves one level deep, assuming there are no terminal "good" states to choose.