DeepMind AlphaGo training tool for joseki Go forum

3 replies. Last post: 2017-12-18

Reply to this topic Return to forum

DeepMind AlphaGo training tool for joseki
  • Carroll at 2017-12-16

    I can’t believe I did not learn about it here:

    https://alphagoteach.deepmind.com/

    Value for black estimated by Master on 10 millions simulations !

  • Carroll at 2017-12-16

    From https://alphagoteach.deepmind.com/dist/files/book.sgf :


    # nodes = 371187

    1. variations = 20991
    2. duplicates = 422

    max path length = 70

    1. human nodes = 8802
    2. missing value = 0
    3. missing ag moves = 0
    4. mirror go positions = 221
    ...

  • Crelo at 2017-12-18

    While I find the AlphaGo games greatly entertaining I do not like this tool so much. The percentage value of the first 23 moves is mostly academic. Possibly interesting for top professionals but hard to use by common players. There are so many things coming afterwards in a go game that a 1.7% difference between the first moves is irrelevant. We really do not know yet why there is this difference and how to actually make it real during the game.

  • Return to forum

    Reply to this topic




    Include game board: [game;id:123456] or [game;id:123456;move:20] or [game;id:123456;move:20;title:some text]