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?

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.

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

Is chess programming about developing a strong chess entity? Then we could just have 2 big projects and everyone works on either side trying to make it the best. Then you can freely use any of their data.

So please tell me, what is chess programming about and WHY should you use other peoples data?

Oh and feeling offended is not always the offenders fault

[Gabor Szots]

Due to the size of the network, we will keep the networks seperate in a submodule of our repository

But it's only 1.5MB while the executables are almost 5MB each. Or I downloaded the wrong net file.

[Gabor Szots]

not wanting Koivisto to be tested by third parties who also list obvious clones which makes the ratings increditable.

I guess it's time you created a list including the obvious clones, of course together with evidence which is clear even for the ignorant. Because rating list people are not experts and you can take it for granted that none of them will study source codes.

[Werner]

Due to the size of the network, we will keep the networks seperate in a submodule of our repository

But it's only 1.5MB while the executables are almost 5MB each. Or I downloaded the wrong net file.

I think you must not download the net. It is excluded in the binaries . And there is no command to set a link to a net.

[Luecx]

Due to the size of the network, we will keep the networks seperate in a submodule of our repository

But it's only 1.5MB while the executables are almost 5MB each. Or I downloaded the wrong net file.

You are right but we also do not store any executable inside the repo. The networks are 1.5Mb but imagine it changes 10 times. The repo size (due to history tracking) will be about 15Mb. People cloning the repo would need to download every single previous network. This way they only get the latest network

[Luecx]

not wanting Koivisto to be tested by third parties who also list obvious clones which makes the ratings increditable.

I guess it's time you created a list including the obvious clones, of course together with evidence which is clear even for the ignorant. Because rating list people are not experts and you can take it for granted that none of them will study source codes.

That’s something I am indeed willing to do

[Luecx]

Due to the size of the network, we will keep the networks seperate in a submodule of our repository

But it's only 1.5MB while the executables are almost 5MB each. Or I downloaded the wrong net file.

I think you must not download the net. It is excluded in the binaries . And there is no command to set a link to a net.

It is included in the binary. Since it is embedded, there is no option to manually set it. Although it’s possible to chose the network to embed while compiling.

[mvanthoor]

I think you missed the point of his posting. If he wanted elo, he would just do what others have done and either used the SF implementation, or at least their data, and almost certainly their trainer. If he just wanted elo, he would have merged the Ethereal net above instead. Instead, he rejected the net even though it was stronger, because it was not HIS data, even though HE trained it and HIS engine runs it.

No, I understood that point perfectly fine, because I try to do the same thing. (At my own development state, obviously.)

What I didn't get is: why implement a neural network if, for whatever reason, you seem to dislike it so much? The implementation did give a boost in playing strength, but implementing a feature despite disliking it, made no sense to me except for the added playing strength.

People will one day regret their gravity towards homogeneity.

To some extent, it is logical. In business software, no-one ever implements anything new if they don't have to. "There is / has to be a library for that" is the often-heard comment if a new feature is requested. Sometimes, I gravitate to libraries as well. I'm totally not interested in finding out how endgame databases work, so I'll probably just use Fathom or Phyrric to access them. I'll probably write some PolyGlot opening book code myself.

[mvanthoor]

So, @Daniel Shawul, we switched to NN to not keep being frustrated by unoriginal engines passing us in the ratings.

rating list position... *shrug*

There are some engines in the rating lists which development started (much) later than mine, but they are already 600 points ahead. To be honest, I don't really care. It all depends on your point of view.

My goal is to try and achieve as high a rating as possible with the least amount of features. Currently, my development version is around 2150 Elo. Some engines in that range already have Null-Move, LMR and other kinds of pruning, staged move generation, and a large amount of evaluation terms. I don't have any of that yet.

Still I'm at the same rating level, so I'm happy. That's the only thing that counts.

Other engines try to get as much playing strength from the least amount of LOC, or even from the least amount of characters possible. Some engines want to be as strong as possible with the smallest executable, or to be as strong as possible while running on a tiny platform such as a micro-controller.

Where other engines are in the rating list doesn't concern me. The only thing that counts is that I feel I got everything from every feature I added, with the cleanest code I'm able to write in Rust at this point.

Is chess programming about developing a strong chess entity? Then we could just have 2 big projects and everyone works on either side trying to make it the best. Then you can freely use any of their data.

When I see such an engine skyrocket through the rating list, it's just.... "There goes another one... onto the ignore list."

So please tell me, what is chess programming about and WHY should you use other peoples data?

For me, chess programming is about learning about the concepts and stuff in computer science I don't use on a daily basis. I use other people's data to 'bootstrap' my engine.

My current development version uses another engine's tapered PST's, so I can see what is at least possible. As soon as I've written my own tuner, I'll probably run it on Zurichess' well-known data-set to see if I can at least match the performance of the PST's of that other engine with my own. Then I'll generate my own data, and try to match or exceed the PST's I made with the Zurichess data.

I know of at least three engines that bootstrapped their first (non-tapered) PST's by using the PST's from my engine because they perform very well in a very basic engine that is just starting out.

I see no problem in using other people's stuff to get started and see what's possible, as long as you replace it with your own before a release.

[Koivisto]

not wanting Koivisto to be tested by third parties who also list obvious clones which makes the ratings increditable.

I guess it's time you created a list including the obvious clones, of course together with evidence which is clear even for the ignorant. Because rating list people are not experts and you can take it for granted that none of them will study source codes.

This is not a valid argument as long as ccrl lists Fat Fritz as #1 on the blitz list.