chess programming • page 2/2 • General Chess Discussion • lichess.org

@radiohead7 well deep neural nets take huge resources. To work and even more to train. If do lightweigth version it is not gonna play well

If you want just experiment take a another game. like four in a row,6x6 othello. Chess, shogi, go will take huge resources

Assuming you want to just learn how this goes and know python:
https://medium.com/applied-data-science/how-to-build-your-own-alphazero-ai-using-python-and-keras-7f664945c188

there are several pyhon projects ans I think there is one generic Zero-style framework where you supply few pluging to main program for rules ans such and will do the neural stuff for you

@radiohead7 well deep neural nets take huge resources. To work and even more to train. If do lightweigth version it is not gonna play well If you want just experiment take a another game. like four in a row,6x6 othello. Chess, shogi, go will take huge resources Assuming you want to just learn how this goes and know python: https://medium.com/applied-data-science/how-to-build-your-own-alphazero-ai-using-python-and-keras-7f664945c188 there are several pyhon projects ans I think there is one generic Zero-style framework where you supply few pluging to main program for rules ans such and will do the neural stuff for you

petri999

#12

@dboing try extract information from other work has limitations. The thing that has provent to be good is to let NN find its way from zero information. AlphaGozero won AlphaGo master 89:11 which is about 450 difference in Elo points. Master won Human master 100 to none so that hard to estimate but we say it not less thatn 800 Elo points stronger than best human. Makaing AlhaGoZeto minimum 1300 points stronger than best humans.

Taking very vague estimate on original AlphaGoLee it was probably about 200-300 points better than Lee and he is top player. And LeeVersion was trained on top human games. That is trying to figure out lessons learned by biological neural nets ove about millenia or half

so at 1100 elopoint for assuming that it best not to assume anything.

Obviousl chess less less Deep game that GO. No one can be 1000+ elopoints stronger than best engines. there will draws enough to stop this

@dboing try extract information from other work has limitations. The thing that has provent to be good is to let NN find its way from zero information. AlphaGozero won AlphaGo master 89:11 which is about 450 difference in Elo points. Master won Human master 100 to none so that hard to estimate but we say it not less thatn 800 Elo points stronger than best human. Makaing AlhaGoZeto minimum 1300 points stronger than best humans. Taking very vague estimate on original AlphaGoLee it was probably about 200-300 points better than Lee and he is top player. And LeeVersion was trained on top human games. That is trying to figure out lessons learned by biological neural nets ove about millenia or half so at 1100 elopoint for assuming that it best not to assume anything. Obviousl chess less less Deep game that GO. No one can be 1000+ elopoints stronger than best engines. there will draws enough to stop this

radiohead7

#13

@dboing @petri999 thank you very much for your info guys. just to make this clear i am not intersted in making an engine playing GOD mode. This is far beyond my computational resources and mostly my skills. I consider a 1600-1800 fide elo engine a huge deal too and if thats achievable im more than willing to take it up and put in the work it requiers. I just dont wanna waste time in some NN that unless it has the most super sophisticated RL algorithm implementation is gonna behave totaly unpredictable and poorly.

wavaxa22 edited

#14

Have you visited http://talkchess.com/forum3/index.php ? Many top notch chess programmers post there.