Gomoku state-of-the-art tech

Question

usually people use pn-search or pn^2 Or df-pn to answer if there is a win solution.

then they use alpha-beta pruning on the min-max game tree with a good evaluation function

they can reach a depth of 15 ply or even more

now there is a Monte Carlo method which is successful in dealing with Go.

Is the same tech can be used in Gomoku ? any examples (source code or paper)

Is there any paper describe a good way to build a well tuned evaluation function.

or Is there any other state-of-the-art or useful tech to deal with Gomoku ?

Is pn search necessary in dealing with Gomoku?

Is there any different VCT engine (src better) ?

Answer 1

To the best of my knowledge, proof number search, dependency based search (also referred as threat space search), and searching algorithms based on alpha-beta framework are mainly used in top Gomoku programs. There also exist some Gomoku programs using Monte Carlo Tree Search, however, the current result is not that good. The article on http://www.aiexp.info/gomoku-renju-resources-an-overview.html summarizes the reading materials, protocols and source code for Gomoku AI.

As for evaluation function, up to now, although there are some papers describing how to build a well tuned evaluation function for Gomoku, none of them really works to achieve the state of the art.

Pn-search is not necessary in dealing with Gomoku. In fact, the state-of-the-art Gomoku engine Yixin does not use pn-search.

Renjusolver is the best VCT engine. Except for renjusolver, there are many other Gomoku engines which have relative good performance on solving VCT and can be downloaded at http://gomocup.org/download/. Currently, pela is the best open source engine on solving VCT.