Solving Hex Openings Using Job-Level UCT Search

Software Development Framework for Job-Level Algorithms

“…In (Wei, Wu, Liang et al, 2014), the JL-Upper Confidence Tree (JL-UCT) algorithm was compared with JL-PNS in the analysis of opening positions for Connect6. In (Liang, Wei and Wu, 2015), several methods to improve JL-UCT were proposed to solve Hex efficiently.…”

“…In contrast, the code for the JL framework requires above ten thousand lines of code. In (Liang et al, 2015), a JL-UCT Hex application was also developed on top of this framework. Lastly, we make concluding remarks in Section 5.…”

“…The second is completely handled by the game-playing programs, and is completely transparent to the BF-JL application, while the third should be carefully handled by JL application developers (role 1). A case for the third is described in (Liang et al, 2015), where all inter-job transpositions are further divided into two sub-classes, according to whether the transpositions are shared among jobs in the same computation node or not.…”

“…The BF-JL searches and games that we have developed are shown in Table 4. Using the framework, we developed a JL application with JL-UCT for Hex in (Liang et al, 2015) by implementing a MoHex handler and using an integrator to integrate both UCT and MoHex handlers together. From this, we can easily develop JL applications with JL-PNS for Go and Hex by simply using integrators to integrate PNS handlers with CGI and MoHex handlers together.…”