Any plans for you and Marc to put piece values in Komodo so people can tweak and tune Komodo, in trying and find the best settings possible ?


Moderator: Ras
The only problem is that some GUIs limit how many UCI options you can have, and I suspect that adding even five (six with bishop pair) basic values for the pieces might exceed some of these limits or at least prevent us from adding new UCI options. We don't object to the idea in principle, it's not a question of secrecy. We do encourage experiments with tuning Dynamism; higher values effectively lower the piece and pawn values (and pawn structure), lower values raise them.
That's good to hear. I figured that was a natural idea to try. I hope you find success with that model. My thought is that not only are LC0 based programs tactically weak, they leave all but a few of the CPU cores inactive. If you can give the CPU your eval function and keep the cores busy doing rollouts at the tree leaves, this could potentially find tactical problems while the MCTS using a slow NN on GPU directs the search. Easier "typed up in a forum post" than done, but the idea has some potential.Well, we have a version of Komodo MCTS that uses a GPU with an NN blended with Komodo MCTS