📝H. G. Muller wrote on Mon, Feb 18, 2019 08:45 AM UTC:
Thanks for warning me about the diagram; the link for the JavaScript powering it was still pointing to my own website rather than to the version here on CVP. Not sure why that did not work; because the file is still there, and I am sure it worked in the past. Perhaps more stringent security measures of modern browsers prevented the access. Anyway, I altered the link to point at the version of the file here on CVP. (The piece images are still taken from my website, though; I guess I should upload these here as well, so I can change that undesirable state of affairs.)
Games with drops are in general harder for computers, relative to humans. Until last year computers still lost to human experts in Crazyhouse. (And then Stockfish-variant appeared.) For regular Shogi the problem was solved earlier, but only through 'data-mining' the huge data base of human Shogi professionals, and absorbing their knowledge in a neural-network type evaluation. For Wa Shogi such a game data base does not exists. And I know only of one computer program that can seriously play it: my own engine Crazywa. (Some non-searching programs like ShogiVar support it, but these are real push-overs.) But I did not seriously tune it even there, and its strategy is simply a generalization of what it does in Crazyhouse, with purely guessed piece values.
I have no doubt that with Google's AlphaZero technology a super-strong player could be made. Wa Shogi is most certainly too insignificant to interest Google, though, or any large community of owners of computers with state-of-the-art graphics cards that would be needed to mimic such an effort (like the LeelaChess Zero project for Chess). So my guess is that computer Wa-Shogi players will remain relatively weak for a long time to come.
Implementing Wa Shogi in Jocly is on my to-do list; it should not be very difficult now that I have already done regular Shogi, so that its infra-structure for drops can be used. Jocly's AI is also laughably weak, however. But as I also made progress in setting up a Jocly based Game Server, it would offer an opportunity to play Wa Shogi on line.
Thanks for warning me about the diagram; the link for the JavaScript powering it was still pointing to my own website rather than to the version here on CVP. Not sure why that did not work; because the file is still there, and I am sure it worked in the past. Perhaps more stringent security measures of modern browsers prevented the access. Anyway, I altered the link to point at the version of the file here on CVP. (The piece images are still taken from my website, though; I guess I should upload these here as well, so I can change that undesirable state of affairs.)
Games with drops are in general harder for computers, relative to humans. Until last year computers still lost to human experts in Crazyhouse. (And then Stockfish-variant appeared.) For regular Shogi the problem was solved earlier, but only through 'data-mining' the huge data base of human Shogi professionals, and absorbing their knowledge in a neural-network type evaluation. For Wa Shogi such a game data base does not exists. And I know only of one computer program that can seriously play it: my own engine Crazywa. (Some non-searching programs like ShogiVar support it, but these are real push-overs.) But I did not seriously tune it even there, and its strategy is simply a generalization of what it does in Crazyhouse, with purely guessed piece values.
I have no doubt that with Google's AlphaZero technology a super-strong player could be made. Wa Shogi is most certainly too insignificant to interest Google, though, or any large community of owners of computers with state-of-the-art graphics cards that would be needed to mimic such an effort (like the LeelaChess Zero project for Chess). So my guess is that computer Wa-Shogi players will remain relatively weak for a long time to come.
Implementing Wa Shogi in Jocly is on my to-do list; it should not be very difficult now that I have already done regular Shogi, so that its infra-structure for drops can be used. Jocly's AI is also laughably weak, however. But as I also made progress in setting up a Jocly based Game Server, it would offer an opportunity to play Wa Shogi on line.