Using Neural Network or Syzygy tables

[OliverBr]

Hello,

now we are watching new engines reaching ELO 3600+ by using Neural Network. A lot of engines are using syzygy tables which is will enhance their performance a lot.

Just asking for a friend:
Is this really the "essence" of chess programming? Aren't the most difficult parts, the eval and the endgame, just being outsourced? Or did every of those engines program those features for their own?

[AndrewGrant]

I'll just chime in at least to say that Syzygy tablebases are worth almost 0 elo now in the top engines.
So that aspect of "outsourc[ing]" is not really relevant.

[Ciekce]

Or did every of those engines program those features for their own?

people aren't sharing networks, despite unfounded accusations from a certain longstanding community member

[OliverBr]

I'll just chime in at least to say that Syzygy tablebases are worth almost 0 elo now in the top engines.

It may be so because most are using Neural Network? Could this render Syzygy tablebases less useful?

So that aspect of "outsourc[ing]" is not really relevant.

So engines could just not use Syzygy tables, but many do. Why?

[OliverBr]

Or did every of those engines program those features for their own?

people aren't sharing networks, despite unfounded accusations from a certain longstanding community member

I am not saying anyone is sharing a network. But did everyone create the program that created the network by him/her/itself?

[mar]

Hello,

now we are watching new engines reaching ELO 3600+ by using Neural Network. A lot of engines are using syzygy tables which is will enhance their performance a lot.

Just asking for a friend:
Is this really the "essence" of chess programming? Aren't the most difficult parts, the eval and the endgame, just being outsourced? Or did every of those engines program those features for their own?

what's the point of all this? are you trying to imply something?!

egtbs only bring a handful of elo at best (I got about 10 self-play with syzygy and my engine is far from top) - so why use it?
I have watched way too many games where the engine was trying to win dead endgame draws, looking silly

as for nets: I use libtorch for low level net optimization, however the trainer itself is written by me.
everything else is completely custom, including topology, indexing scheme and inference code
there are also people who wrote trainers completely from scratch without even relying on third party libraries
also I only used self-play for training, no external factors involved except data was generated with syzygy enabled
and btw generating positions is by far the most time consuming process

lastly I'd like to thank Jay and Peter again, who were very helpful when I was transitioning my engine to NNs

we are all "outsourcing" compilers and operating systems. even compilers often "outsource" LLVM

[towforce]

I'll just chime in at least to say that Syzygy tablebases are worth almost 0 elo now in the top engines.
So that aspect of "outsourc[ing]" is not really relevant.

Thank you: would not have realised that if you hadn't "chimed in".

[OliverBr]

I'll just chime in at least to say that Syzygy tablebases are worth almost 0 elo now in the top engines.
So that aspect of "outsourc[ing]" is not really relevant.

Thank you: would not have realised that if you hadn't "chimed in".

It may be so for "Top" engines, especially with a neural net.
But: If you consider a "normal" engine, everything hand-crafted, own ideas, no neural network, I really do think, that Syzygy tablebases increase the ELO of such an engine.

[lithander]

Tapered PSQTs where you find the the values with something like Texel's tuning method is considered normal and an essential part of a basic HCE, right?

But it's not more or less original and handcrafted than using NNUEs. In both cases the chess knowledge is encoded in weights/coefficients and found via machine learning approaches.