一栗仆直/Ranar

從 Luzi82_wiki

跳到: 導航, 搜尋

目錄

主旨

用 Java 寫一個 Genetic Programming (GP) 的過三關 AI ,純粹用作研究 GP 效能之用。

Roadmap

Ranar 0 (Evolution)

  • To have an GP-AI able to decide move in a game.
  • Each candidate would be ranked by the followings:
    • Number of match won. And how quick it was.
    • Number of match lost. And how long it surival.
    • Number of successful move made. That mean, it would not mark on a cell which is already filled.

Ranar 0.01 (Applet)

  • Have a Ranar Engine run as an Applet, backed by at evolution pool running on the server.

Ranar 1 (Recursion)

  • Instead of deciding move, the AI is now have to rank all available move in the game, for alpha-beta searching optimization.
  • The AI have to decide:
    • The order of searching.
    • Which move to be pruned.
    • Node rating.

Ranar 2 (C)

How to obtain

svn co svn://luzi82.servehttp.com/wings/TicTacToe_GP
個人工具