Maybe it would be interesting to learn what the networks are doing that the hand tuned evaluations are not doing.AndrewGrant wrote: ↑Fri Oct 09, 2020 11:22 am An hour after I make this post I'm filled with regret.
[1] Do people really want to see a bunch of NNUE engines, all running the same evaluation function, occupying the top slots of computer chess?
There must be a reason that they are tearing everyone's heads off and playing soccer with them. Maybe if we started doing the important things we are overlooking, we would not see the enormous gap.
As far as the same evaluation function, every net seems to create different results. There must be a reason for that too.
It is interesting if you hand tune. It is interesting if you use a net. Why not understand what the nets are doing and make a better way of net creation? Perhaps there is a way to use the hardware better. Perhaps there is a way to train them better.[2] Would people really find it interesting to have Ethereal use whatever "best" network is out there?
I like weak engines. I like strong engines. I like hand tuned engines. I like programatically tuned engines.[3] Are people still interested when engines replace 50% of what made them unique?
Ethereal is a very interesting project. I like the code. I like the engine. I don't care if you ever pass LC0 and Stockfish. I will like Ethereal the same whether it becomes the strongest engine in the world or it does not do that.[4] Do people want to continue to have Ethereal, acting as a Stockfish-lite, for new programmers?
I do not think it is bad, what is happening with Shogi and Go. Both of those games are much harder than chess and they invented this incredible NNUE stuff. I do not dislike NNUE, but I also do not think it is necessary to use it if you don't want to.[5] Is Chess doomed to go the route of Shogi and Go? We share the Networks, but have our own searches? Is that okay?
I am OK with you doing anything that you like. If you want to pursue eval, go ahead. I think your efforts are better spent on search if you want to win, but if you are pursuing art and beauty then that is probably irrelevant.[6] Is this chess world okay with me plunking in a Network file and never changing it again?
If someone can convince me of all of those things, or at least most of them, then maybe there is still work to be done. The OCD in me likes the idea of refactoring all of the garbage in the NNUE libs. Likes to idea of coming up with a cleaner solution than what I did to hack the code in with such speed. I'de have to look at benchmarks. If NNUE is the "great equalizer", then its of little substance. But if NNUE is +160 elo flat across the board, then maybe its not the end of the world.
At the same time, NOT quitting goes against everything I've preached in this forum for the last few years. I don't even like people using Ethereal data to tune their engines. Alayan and I were going to release datasets to go with my tuning paper, but I did not like the idea of people converging to Ethereal. But now its okay that I drag and drop 2,000 lines from Stockfish? Its okay that I just run whatever Network Fishtest has brewing or Dkappe has brewing or someone else?
I'm not a net trainer. I won't do it, its not interesting. Ethereal's tuner has lots of things akin to net training. Lots of knobs to tweak. but it is a means to and ends, and that ends is putting human chess knowledge into the engine. The ends of Network training is to get a "better" black box.
I don't know if I'm ripping myself away from something I love for no reason. If I'm doing it because I see the pain in the future. If I'm doing it because I feel the pain now but am too daft to see the light at the end of the tunnel. I need someone to convince me that this hobby is still worth pursuing. That beating Leela and Allie and Scorpio is worth something. That doing so is not just a result of NNUE, but also a result of the ten thousand hours I've put into the project. That I'm not sacrificing my ideals and my principles by continuing on in this new era.