Koivisto 5.0

[Luecx]

So, @Daniel Shawul, we switched to NN to not keep being frustrated by unoriginal engines passing us in the ratings. do you think that using other peoples data, or anything else out of the 3 i listed is actually the right thing to do?

So your engine's postion on the rating lists is all you care about, what a surprize! That is hardly a position to be to lecture others on the value of "real men evaluation". You are saying those who use NNUE backends (or trainers) are "unoriginal" while you yourself now uses it. That you re-invented the wheel, wrote trainer code from scratch!!, gets you zero (and maybe even negative) points.

What is the point in making an engine based on using other peoples code. I am not talking about pruning ideas here but about replacing 1/2 of the engine by other peoples code.

Using library code is not a crime. Using free data is not a crime.

So what is chess programming about? Just making a strong engine ..

That definately is the case for you...

and trying to catch up to the very top by just taking their code?
If that's the case, I will admit that I am wrong with everything and people should just continue using other peoples stuff... If it is about making a strong engine without such, you need to put a lot more effort into an engine than everybody else and simply being passed their by those who do take other peoples code feels depressing.

It is hypocritical to lecture others while doing the same thing as what they are doing. You spend half your post on your engine's release ranting about how other engines are un-original, how they ctrl+c/ctrl+v Stockfish code etc. You must have the most original engine in the universe with no code written by others, heh? My guess is you have an engine that shot up to the rating lists with ideas taken from stockfish, and now you feel disappointed others caught up with you using NNUE, and now you say look at me I wrote trainer from scratch to justifiy using NNUE and massage your ego! You are not the first guy to feel that way...

It is funny how you only read what you want to read. I never said we only care about the ranking. I explicitely said we do not. Otherwise we would have taken Eth-Data net. Using other peoples data and training tools and whatever is not theft but selling the engine as original is just hilarious...

[Daniel Shawul]

It is hypocritical to lecture others while doing the same thing as what they are doing. You spend half your post on your engine's release ranting about how other engines are un-original, how they ctrl+c/ctrl+v Stockfish code etc. You must have the most original engine in the universe with no code written by others, heh? My guess is you have an engine that shot up to the rating lists with ideas taken from stockfish, and now you feel disappointed others caught up with you using NNUE, and now you say look at me I wrote trainer from scratch to justifiy using NNUE and massage your ego! You are not the first guy to feel that way...

I interpreted that "same thing" to be ctrl+c/ctrl+v code from Stockfish. What "same thing" are you referring to?

Unbelievable, you just apologized to me that you misunderstood what I said!
Now you go back to saying I said he copied code, which is it ? Here a recap for you since you seem to have a problem comprehending posts.
He said:

The easiest part is probably the NN implementation inside the engines themself although even here many people seem to ctrl+c, ctrl+v popular implementations.

What I said is hypocritical is him lecturing others about how HCE is a "real man's evaluation", while at the same time implementing NN
for a quick +300 elo again. The justification he provided is that he wrote trainer from scratch (is it in C too lol) ...

[Koivisto]

It is hypocritical to lecture others while doing the same thing as what they are doing. You spend half your post on your engine's release ranting about how other engines are un-original, how they ctrl+c/ctrl+v Stockfish code etc. You must have the most original engine in the universe with no code written by others, heh? My guess is you have an engine that shot up to the rating lists with ideas taken from stockfish, and now you feel disappointed others caught up with you using NNUE, and now you say look at me I wrote trainer from scratch to justifiy using NNUE and massage your ego! You are not the first guy to feel that way...

I interpreted that "same thing" to be ctrl+c/ctrl+v code from Stockfish. What "same thing" are you referring to?

Unbelievable, you just apologized to me that you misunderstood what I said!
Now you go back to saying I said he copied code, which is it ? Here a recap for you since you seem to have a problem comprehending posts.
He said:

The easiest part is probably the NN implementation inside the engines themself although even here many people seem to ctrl+c, ctrl+v popular implementations.

What I said is hypocritical is him lecturing others about how HCE is a "real man's evaluation", while at the same time implementing NN
for a quick +300 elo again. The justification he provided is that he wrote trainer from scratch (is it in C too lol) ...

Most of the gain between 4.0 is in search and hce, only about 100 stc self play elo comes from the NN. Infact we had similar gains from search and hce since tcec version that we had from NN.

[connor_mcmonigle]

It is hypocritical to lecture others while doing the same thing as what they are doing. You spend half your post on your engine's release ranting about how other engines are un-original, how they ctrl+c/ctrl+v Stockfish code etc. You must have the most original engine in the universe with no code written by others, heh? My guess is you have an engine that shot up to the rating lists with ideas taken from stockfish, and now you feel disappointed others caught up with you using NNUE, and now you say look at me I wrote trainer from scratch to justifiy using NNUE and massage your ego! You are not the first guy to feel that way...

I interpreted that "same thing" to be ctrl+c/ctrl+v code from Stockfish. What "same thing" are you referring to?

Unbelievable, you just apologized to me that you misunderstood what I said!
Now you go back to saying I said he copied code, which is it ? Here a recap for you since you seem to have a problem comprehending posts.
He said:

The easiest part is probably the NN implementation inside the engines themself although even here many people seem to ctrl+c, ctrl+v popular implementations.

What I said is hypocritical is him lecturing others about how HCE is a "real man's evaluation", while at the same time implementing NN
for a quick +300 elo again. The justification he provided is that he wrote trainer from scratch (is it in C too lol) ...

Your post was ambiguous. I acknowledged that it was possible I misinterpreted your post and apologized. I then asked for clarification relating to the ambiguity indicated in my previous post.

Anyways, his justification is that his implementation, training code (https://github.com/Luecx/Differentiatio ... master/src), training data, inference code etc. are all not copied from another engine. He's not doing the "same thing" because he's not copy pasting Stockfish's implementation.

[Madeleine Birchfield]

Using neural networks does not require any understanding of chess which you would need when writing a hand-crafted-evaluation or as we like to refer to: real-men-evaluation (RME). Using neural networks became more of an engineering challenge than anything else.

Neither does using and tuning Piece-Square Tables in your evaluation function require any understanding of chess, any more than NNUE's king-piece square tables require understanding of chess.

[Luecx]

Using neural networks does not require any understanding of chess which you would need when writing a hand-crafted-evaluation or as we like to refer to: real-men-evaluation (RME). Using neural networks became more of an engineering challenge than anything else.

Neither does using and tuning Piece-Square Tables in your evaluation function require any understanding of chess, any more than NNUE's king-piece square tables require understanding of chess.

You cant find a RME beyond 3000 Elo which utilizes no chess specific code beyond psqt. So what are we talking about? I am talking about simple stuff like evaluating the bishop pair and more complex stuff like outposts, connected pawns, king safety etc.

[Daniel Shawul]

Hand-crafted evaluation is now more like "Cave Mens evaluation" than "Real Mens evaluation"

[mvanthoor]

Hand-crafted evaluation is now more like "Cave Mens evaluation" than "Real Mens evaluation"

To be honest, even HCE not completely HCE anymore. Now, we make a table that gives bonus/malus for open lines, doubled pawns, etc, and we have Texel Tuning tune those. It's like a poor man's neural network, where you define each parameter to be tuned yourself. A neural network just has a huge amount of parameters with no name. The "hand-crafted" part is only assigning when some parameters should be used. ("If you have an open line, use that value. If you have doubled pawns, use this value.") That's not really hard to do. The big advantage of a HCE is that humans can clearly understand what the engine is doing, and by looking at the games, you can clearly see the impact of a newly added feature.

[Madeleine Birchfield]

Using neural networks does not require any understanding of chess which you would need when writing a hand-crafted-evaluation or as we like to refer to: real-men-evaluation (RME). Using neural networks became more of an engineering challenge than anything else.

Neither does using and tuning Piece-Square Tables in your evaluation function require any understanding of chess, any more than NNUE's king-piece square tables require understanding of chess.

You cant find a RME beyond 3000 Elo which utilizes no chess specific code beyond psqt. So what are we talking about? I am talking about simple stuff like evaluating the bishop pair and more complex stuff like outposts, connected pawns, king safety etc.

What does elo have anything to do with your claim that neural networks do not require any understanding of chess and therefore any engine using neural networks does not have a real man evaluation? I don't care about elo.

Many engines using neural networks have handcrafted features such as king safety and bishop pair as inputs into their neural networks, such as Wasp and Winter (and KataGo with ladders in computer Go), and others use neural networks as only part of their evaluation and handcrafted stuff for the rest, such as Fizbo and Ethereal 12.75. Saying that Wasp and Winter and Fizbo and Ethereal 12.75 (and Stockfish for that matter with its hybrid eval) do not have 'real man evaluations' because they use neural networks does not make any sense. By that logic, any engine using piece-square tables does not have a 'real man evaluation' because piece square tables do not require chess knowledge beyond what is needed for understanding nnue's king piece square tables.

[Daniel Shawul]

Hand-crafted evaluation is now more like "Cave Mens evaluation" than "Real Mens evaluation"

To be honest, even HCE not completely HCE anymore. Now, we make a table that gives bonus/malus for open lines, doubled pawns, etc, and we have Texel Tuning tune those. It's like a poor man's neural network, where you define each parameter to be tuned yourself. A neural network just has a huge amount of parameters with no name. The "hand-crafted" part is only assigning when some parameters should be used. ("If you have an open line, use that value. If you have doubled pawns, use this value.") That's not really hard to do. The big advantage of a HCE is that humans can clearly understand what the engine is doing, and by looking at the games, you can clearly see the impact of a newly added feature.

Absolutely! HCE is just a perceptron with a linear activation function. "Texel tuning" is just training a perceptron with sigmoid activation function so as to map it to game results. So basically HCE is humans hand-selecting the weights and features, i.e. if you don't use a tuner, and if you do use tuner the human's job reduces to selecting features. If hand-selecting evaluation features is a "real man's" job, I will pass ...
Even Komodo, with a Grand master in its team, have moved to NNUEs now.