Re: New engine releases 2020
Posted: Wed Nov 25, 2020 2:19 pm
Of course you can :) I was too lazy to announce Rubi, but my site is up to date.Frank Quisinsky wrote: ↑Thu Nov 26, 2020 2:28 pm Hi Guenther,
not sure I can use the thread if I see new versions are available.
If not ... sorry for my message!
Best
Frank
What's the deal with Rubi's NNUE implementation? Is it "original" or another SF clone? I see it's disabled by default, and the user has to manually enable it to use the networks. When testing, should we leave it disabled (like Minic) for now?Frank Quisinsky wrote: ↑Thu Nov 26, 2020 2:23 pm Hi there,
1. RubiChess
now RubiChess 1.9 NNUE is available.
Very nice for my next FCP Tourney-2021.
Have many thanks Andreas!
And sorry for 1-2 messages from myself in your RubiChess thread (nothing to do RubiChess).
The NET file is inside the download file!
https://github.com/Matthies/RubiChess/
But here in special for NET files!
https://github.com/Matthies/NN
Very clearly arranged, easy to install.
Andreas prefers NNUE as default.CMCanavessi wrote: ↑Fri Nov 27, 2020 1:25 amWhat's the deal with Rubi's NNUE implementation? Is it "original" or another SF clone? I see it's disabled by default, and the user has to manually enable it to use the networks. When testing, should we leave it disabled (like Minic) for now?
My NNUE net is 100% based on my own evaluations training data.
Yeah, but that's just the network. What about the implementation, trainer, etc?Graham Banks wrote: ↑Fri Nov 27, 2020 3:10 amAndreas prefers NNUE as default.CMCanavessi wrote: ↑Fri Nov 27, 2020 1:25 amWhat's the deal with Rubi's NNUE implementation? Is it "original" or another SF clone? I see it's disabled by default, and the user has to manually enable it to use the networks. When testing, should we leave it disabled (like Minic) for now?
The NNUE is his own:My NNUE net is 100% based on my own evaluations training data.
Implementation is SF per the source code.CMCanavessi wrote: ↑Fri Nov 27, 2020 4:09 amYeah, but that's just the network. What about the implementation, trainer, etc?Graham Banks wrote: ↑Fri Nov 27, 2020 3:10 amAndreas prefers NNUE as default.CMCanavessi wrote: ↑Fri Nov 27, 2020 1:25 amWhat's the deal with Rubi's NNUE implementation? Is it "original" or another SF clone? I see it's disabled by default, and the user has to manually enable it to use the networks. When testing, should we leave it disabled (like Minic) for now?
The NNUE is his own:My NNUE net is 100% based on my own evaluations training data.
Then why is 'use NNUE ' false by default?