LC0 has a great disadvantage against weaker strategical knowledge, as we know well.
This mainly arises from its structure and working method. But there is an another cause: namely the training with self playing. It is an old true but real now also to learn most faster and most more you need a cleverer teacher than you are.
So moreover the self playing, the developer of Leela should use classical engines with AB searching to train Leela.
Maybe the positional knowledge of Leela would be weaker in some measure but her tactical knowledge can enhances drastically.
Leela Chess 1/2 would be stronger than LC0
Moderators: hgm, Rebel, chrisw
-
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
-
- Posts: 266
- Joined: Fri Jul 10, 2015 9:23 pm
- Location: Russia
Re: Leela Chess 1/2 would be stronger than LC0
Eugene Kotlov
Hedgehog 2.1 64-bit coming soon...
Hedgehog 2.1 64-bit coming soon...
-
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: Leela Chess 1/2 would be stronger than LC0
Why not?
There is no any evidence against my sentence.
But basing on common sense it is very likely it will prove to be true.
There is no any evidence against my sentence.
But basing on common sense it is very likely it will prove to be true.
-
- Posts: 1142
- Joined: Thu Dec 28, 2017 4:06 pm
- Location: Argentina
Re: Leela Chess 1/2 would be stronger than LC0
Yes there is. Compare Alpha Go with Alpha Go Zero and Alpha Go Lee...
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls
-
- Posts: 4190
- Joined: Wed Nov 25, 2009 1:47 am
Re: Leela Chess 1/2 would be stronger than LC0
Apples and oranges. Why don't you compare it to SC2LE, it makes equal (lack of) sense?CMCanavessi wrote: ↑Tue May 08, 2018 3:52 pmYes there is. Compare Alpha Go with Alpha Go Zero and Alpha Go Lee...
-
- Posts: 362
- Joined: Thu Mar 16, 2006 7:39 pm
- Location: Portugal
- Full name: Alvaro Cardoso
Re: Leela Chess 1/2 would be stronger than LC0
The Lella Chess project is trying to "replicate" the Alpha Zero project.
The Alpha Zero project is based on very different concepts from classical AB engines and they did not use AB engines to train the networks.
Remember the A0 team isn't stupid nor ignorant, they are pretty smart.
The A0 team proved (at least to me) that their methods really works, you just need to implement them very well.
The LC0 project is also proving these methods works. The main problem of LC0 is in my opinion the hardware. The Google team has access to fabulous hardware while the LC0 team don't, so they will have to wait quite a bit longer to train the networks. Unfortunately also during game play it is very limited in nodes per second, maybe in the future someone will aqquire some of those TPUS and test LC0 on that hardware. While you can run a classical AB engine on your PC and get very powerfull play from it you won't be able to get high performance from LC0 on a "normal" PC, at least on mine with a Radeon 4600 HD and a core i7 4 core Nehalem I won't get anywhere with LC0.
The Alpha Zero project is based on very different concepts from classical AB engines and they did not use AB engines to train the networks.
Remember the A0 team isn't stupid nor ignorant, they are pretty smart.
The A0 team proved (at least to me) that their methods really works, you just need to implement them very well.
The LC0 project is also proving these methods works. The main problem of LC0 is in my opinion the hardware. The Google team has access to fabulous hardware while the LC0 team don't, so they will have to wait quite a bit longer to train the networks. Unfortunately also during game play it is very limited in nodes per second, maybe in the future someone will aqquire some of those TPUS and test LC0 on that hardware. While you can run a classical AB engine on your PC and get very powerfull play from it you won't be able to get high performance from LC0 on a "normal" PC, at least on mine with a Radeon 4600 HD and a core i7 4 core Nehalem I won't get anywhere with LC0.
-
- Posts: 143
- Joined: Wed Jan 17, 2018 1:26 pm
Re: Leela Chess 1/2 would be stronger than LC0
What exactly would we need a neural network engine for if we cripple it from the start? Using AB engine self-play cannot possibly help to make Leela stronger, but it would destroy the distinctive, positional playing style that chess players who follow this project actually appreciate.corres wrote: ↑Tue May 08, 2018 1:18 pm LC0 has a great disadvantage against weaker strategical knowledge, as we know well.
This mainly arises from its structure and working method. But there is an another cause: namely the training with self playing. It is an old true but real now also to learn most faster and most more you need a cleverer teacher than you are.
So moreover the self playing, the developer of Leela should use classical engines with AB searching to train Leela.
Maybe the positional knowledge of Leela would be weaker in some measure but her tactical knowledge can enhances drastically.
Self-play should remain what it is, self-play. This does not mean that Leela cannot at some point be coupled with an Alpha-Beta engine to strengthen its tactics while leaving its positional power intact. But this should be done in a way that it is optional, at the user side (for competitions etc.), and can still be deactivated if people prefer that positional, but tactically vulnerable Leela e.g. as sparring partner.
-
- Posts: 4190
- Joined: Wed Nov 25, 2009 1:47 am
Re: Leela Chess 1/2 would be stronger than LC0
The problem is the following. AG0 like AG is based on MCTS which was the only successful concept in Go. So there was nothing new (at least in AG case) except monstrous hardware. With AG0 they introduced reinforcement learning. This is also fine, because their game model is Go where we have known for quite a while (well before Google got its hands on it) that MCTS+NN approach works. Additionally to reinforcement learning, in AG0 they unified policy and value network in one as another novelty. And we can say this has been reasonably demonstrated that it works for Go.Cardoso wrote: ↑Tue May 08, 2018 4:54 pm The Lella Chess project is trying to "replicate" the Alpha Zero project.
The Alpha Zero project is based on very different concepts from classical AB engines and they did not use AB engines to train the networks.
Remember the A0 team isn't stupid nor ignorant, they are pretty smart.
The A0 team proved (at least to me) that their methods really works, you just need to implement them very well.
The LC0 project is also proving these methods works. The main problem of LC0 is in my opinion the hardware. The Google team has access to fabulous hardware while the LC0 team don't, so they will have to wait quite a bit longer to train the networks. Unfortunately also during game play it is very limited in nodes per second, maybe in the future someone will aqquire some of those TPUS and test LC0 on that hardware. While you can run a classical AB engine on your PC and get very powerfull play from it you won't be able to get high performance from LC0 on a "normal" PC, at least on mine with a Radeon 4600 HD and a core i7 4 core Nehalem I won't get anywhere with LC0.
Chess is a different game altogether and just blindly assuming the concepts from Go would work equally well without proof is plain wrong.
Some take that lousy preprint as a proof, some are intimidated by Google and would take whatever they say at face value, fine, but I disagree with that.
We don't know if MCTS (or UCT in what ever version) would work for chess (there are strong indications it doesn't). We don't know if combining policy and value network in one NN would work for chess (so far judging by atrocious LC0 tactical performance it doesn't).
Yes LC0 is based on L0 simply because A0 is based on AG0. And what we know so far (beside awful tactical performance and very strong positional performance of LC0):
1) Training is saturated (it is a current state no matter how much some ppl bury their head in send). Whether it is because of bug in training code, small network size, bad training process, or simply too shallow training games due to the lack of monstrous hardware that Google had at their disposal is to be seen.
2) Despite having relatively large net (half of the A0) performance based on extrapolation to equal hardware as A0 (4TPUs) is 300-400Elo below A0. And we know that increasing net size brings initially very little if anything because of much slower evaluation. Doubling net size usually slows down nps for factor 2.5-3. When going from 6x64 to 10x128 Lc0 gained a lot. When going from 10x128 to 15x192 much, much less, and I'm talking here about testing with fixed number of playouts.
3) Scaling performance is not nearly as good as some expected it would be (ofc I would say) and is totally different to what has been presented in that lousy preprint. It is 50-60 Elo per doubling of playouts.
-
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: Leela Chess 1/2 would be stronger than LC0
I think there are at least three causes why the developers of Leela Chess should not stick to self-play training only:
1, To learn tactics from a tactically weak chess machine is a very unproductive thing
2, The developers have most lesser resources to develop Leela than DeepMind have
3, They want to make a super strong chess player entity only - in contrast to DeepMind who want to prove their method used for AlphaZero is an universal method for formation recognizing applications using NN technique.
Because of this DeepMind uses self-play training only
1, To learn tactics from a tactically weak chess machine is a very unproductive thing
2, The developers have most lesser resources to develop Leela than DeepMind have
3, They want to make a super strong chess player entity only - in contrast to DeepMind who want to prove their method used for AlphaZero is an universal method for formation recognizing applications using NN technique.
Because of this DeepMind uses self-play training only
-
- Posts: 1470
- Joined: Mon Apr 23, 2018 7:54 am
Re: Leela Chess 1/2 would be stronger than LC0
1. We'll find out. Can you be 100% sure?corres wrote: ↑Tue May 08, 2018 6:14 pm I think there are at least three causes why the developers of Leela Chess should not stick to self-play training only:
1, To learn tactics from a tactically weak chess machine is a very unproductive thing
2, The developers have most lesser resources to develop Leela than DeepMind have
3, They want to make a super strong chess player entity only ...
2. The community has more time & patience & no commercial/publicity cravings.
3. Who said? Weaker but pure is okay for many of us.