If you are a correspondence player, you need Leela now! - A corr GM opinion on Leela

Sort:
drmrboss
MyGreatMethod1 wrote:

 I was mistaken that it is weaker than Fritz 5, Weaker than Fritz 4 if its only like 2400. I never tried  Leela UCI installing to PC because it is clear sales talk. They are planning to go commercial soon most probably.

Leela is free and open source (community driven project with GPL 3.0, similar to Stockfish project).

 

Leela will never go commercial! ( you cant make commercial with GPL 3.0 )

 

https://github.com/LeelaChessZero

 

P.S, As of today, there are a dozen of developers and 100+ contributors who are donating their GPU . (I am one of GPU contributor) 

drmrboss

http://talkchess.com/forum3/viewtopic.php?f=2&t=68621

Read this, of course Leela is about -50 elo weaker than SF in fair hardware, but Houdini and Komodo were about -70 elo weaker than SF. 

What do you mean by very weak?

There are lots of tests in Lco forum where $1200 2080ti vs $1300 16 cores CPU  SF(Leela is about -30 to -50 elo below SF).

Search yourself. But if you think Leela is 2400 elo you should go out of the cave and read the news.  wink.pngcry.pngshock.png

 

drmrboss

https://docs.google.com/spreadsheets/d/1XSJiCcQpCLv0fNwrUn7jXjdkZFU63YFEWpdXv6dSSg0/edit#gid=1707964751

P.S There are also people who never heard about stockfish and they think Fritz/ Rybka is the strongest. It is not my interest to convince such people with ignorance. 

EscherehcsE
EscherehcsE wrote:

I just revisited Leela Chess Zero (ver 0.18.1, network #11250) on my cheap hardware (no GPU, one thread). It's getting better - Up to about 2400, holding its own with Dragon 4.6. Now with multi-PV and Syzygy tablebase support, it might be worth considering as a supplemental analysis engine to find variations that the stronger, more highly pruned engines might miss.

After tweaking the config settings, I was able to greatly reduce the number of time forfeits - Now the time forfeits are only about one in a hundred games. On my crappy PC (without a GPU), Leela is now playing at a 2500+ level. Not bad for no GPU!

Tweaked settings:

Scale thinking time: 0.500000
Move time overhead in milliseconds: 10000

EscherehcsE

@MyGreatMethod1 - You don't seem to understand that you're comparing apples to oranges. The CCRL ratings for Leela are made using hardware WITHOUT a GPU. The discussions by drmrboss, stating that Leela is close to the top engines, are based on Leela using hardware WITH a GPU.

drmrboss
MyGreatMethod1 wrote:

While latest Stockfish is rated 3459 on your PC. Leela not even close to top 30. Since CCRL rating is inflated, let's say Leela strength is close with Fritz5's strength. Fritz 5 is stronger than Deep Blue not bad achievement for Leela.

 Looks like you are quite naive about Leela.

null

https://www.chess.com/computer-chess-championship

 

Leela  start with "zero " knowledge or nearly zero rating. That ccrl tests were done in early version of Leela chess zero in about 4 months ago, . (which is no longer in development anymore). .If you use earlier version of Leela, her rating may be lower than chimpanzee. Lol. You can ask CCRL organizer Graham Bank when they tested that Leela version. (gbanksnz at gmail.com)

The one I am talking is Lco, which run on GPU with CUDA support. (sorry for technical term to someone who is naive about Leela (who dont even know whether it is free or commercial engine).

drmrboss

Read the orignial post, correspondence GM is using GTX 2080 

Looks like you dont know what is GPU or what is GTX  2080 is.

Here is a link, learn those first, then you should talk, otherwise it is too hilarious !!

https://www.nvidia.com/en-au/geforce/graphics-cards/rtx-2080/

drmrboss
MyGreatMethod1 wrote:

Did you try it  on your laptop Fritz 6  badly punish Leela. Hahahaha. Do not give me marketing nonsense just try it on your PC.

 

No need to mention it again, I alreday know your limited knowlege about engines.  

You should not be commenting on  this topic, my topics are for those enthuasitc chess players/ correspondence players/ who already using engines for chess.  

drmrboss

It is not about Leela's problem, it is your problem in limited knowledge.

There are also people who dont know how many cores of SF is running, how many depth of SF is running etc. (  They claim they beat SF).

Titled_Patzer


Taskinen

13 days ago

#13
Johnny_Gumshoe wrote:
I did not know it was allowed and quite frankly alarmed that it is.

Taskinen: : "In chess.com daily games using engine to evaluate moves is not allowed. But in serious rated correspondence games, engine use is allowed."

I take serious difference to this statement and a previous one from his past post.

1st. Engine use is allowed at chess sites that explicitly permit it's use. Chess.com does Not permit engine use in correspondence chess.

2nd. To report that the detection thereof is impossible to detect only further encourages it use. (From a previous post of Taskinen. 

3. Correspondence players since the time of post cards have been playing "serious" chess. Most players are above board. More so than online live players. From my experience, correspondence players exhibit a high level of respect and honesty. 

It is unfairly and incorrectly suggested, that longer games, specifically daily chess, lends itself to a greater % of cheaters. (Because of the belief it is harder to detect, therefor people become "prone" to cheating.) This is an incorrect assumption. Cheating is a matter of character. It ill done by those so inclined, regardless of the time control. (Many still do not realize the easy methods to cheat at bullet. The fact is cheating in bullet is as common as any time control.)

I am reading today in chat at another site covering the Isle of Man that 250 Titled players at CC have been found to be cheating, in events as Titled Tuesday. Can't verify this as yet, but it was common knowledge in the chat. Point is, cheating is found everywhere. Less so in correspondence IMO because those players take more "pride" in their daily games.

The OP is suggesting correspondence players need to get Leela. Why? To use for analysis of ongoing games ? This is a great disservice.

Titled_Patzer

The OP is a prominent and active member of two "Trolling" clubs. Makes the claim he is a corr GM. to which no evidence is given and is a Rarity.

No further bio info is needed in defining his character. IMO, he is "promoting" the use of engines in correspondence play, encouraging cheating at CC.

 

drmrboss

@Titile Pazter, The orignial post I copied from ICCF GM who posted in our leela. You have to aware that correspondence chess doent only mean " your daily chess chess,com". The most prestigious correspondence chess is ICCF., you are allowed to use any assistance.

 

 

drmrboss
Titled_Patzer wrote:

The OP is a prominent and active member of two "Trolling" clubs. Makes the claim he is a corr GM. to which no evidence is given and is a Rarity.

No further bio info is needed in defining his character. IMO, he is "promoting" the use of engines in correspondence play, encouraging cheating at CC.

 

You go and check leela blog on that date by yoursefl.

https://blog.lczero.org/

Check his rating on FIDE and correspondence chess. ( It is not my post, he is FIDE FM, not sure his GM status in ICCF though according to his claim).

drmrboss

Different people have different opinion on Leela. In this game , Leela won vs Stockfish and Kingscrusher say" End of era near for classical engines, Stockfish" grin.png 

https://www.youtube.com/watch?v=N4WCsRvBrtc

congrandolor

Cheating in bullet is stupid, shut up please.

superchessmachine
Daniel1115 wrote:

 

LouStule wrote:

Agreed. What’s the point of playing if everyone is using an engine? Bragging rights I guess “My computer is smarter than your computer”.

My computer has more Ram than yours. Alternatively:I have no life and can run my engine the whole day every day(or whatever the interval is)

 

How many GB you got boi?

drmrboss
MyGreatMethod1 wrote:

I always tell you try Leela it on your own PC. It is as hoax as Google Alpha Zero. You cannot try Alpha Zero in your PC though.

Another limited information again. A0 won vs limited version of Stockfish 8, by 28 win 72 draws, that means A0 is only +100 elo above that limited version of SF 8. Current Leela is playing against Full strength SF Developmental version with opening books and endgame table bases ( Stockfish 10) which is more than +100 elo stronger than SF 8.

Leela support Endgame Tablebase, A0 did not.

Tests shows that A0 and Leela are within 50-100 elo .( In favour of A0)

A0 speed is 80Knps, Current Leela on 2x2080 Ti is 50-60 Knps. 

 

They are very close now.

 

Dont get fool with google adverting( A0 is advertisment only, quite similar to how to be a millionarie in 1 week etc).

 

Leela is exact replica of Alpha Zero currently run by exactly the same Neural network structure and learning methods 

 

Since a few months ago, developers are adding more and more better learning methods than the technology they used in 2017 for A0.

 

Remember, today technology is " year 2018". Neural Network is advancing in very fast pace.

drmrboss
congrandolor wrote:

Cheating in bullet is stupid, shut up please.

No, they wont. I have been accused as cheater in bullet several times. ( chess.com and lichess).

There are still a lot of people who always think their loss is due to encountering with cheaters. ( and only a few people express their mind/accuse you as cheater, lol).

drmrboss

 

I told you several times. Dont be stubborn, learn yourself. 

This is an official leela tester's rating from a home computer. (see his setting)

I see leela on GTX 980ti =3488 , 4 cpu SF 9 3560 (SF 10 should be 3610) , 4 cpu Fritz 16 (Rybka 5) is only 3232( Rybka 3 would be 3150 approx)

https://docs.google.com/spreadsheets/d/1XSJiCcQpCLv0fNwrUn7jXjdkZFU63YFEWpdXv6dSSg0/edit#gid=1707964751

 

Again, it is practically and technically very difficult to connect two computers from two countries. Not surprising, you have very limited knowledge about networking as well.  (Some network programmers use public API, but it is not my interest in this topic) 

 

This is benchmark of Leela testers, to get a standardization.  You can see my name on row 10, Nay Tun. 

https://lc0bench.netlify.com/bench-result.html

Well, I am doing advertising about Leela since the beginning of Leela project. 

https://www.chess.com/forum/view/general/leela-zero-a-neural-network-engine-similar-to-alpha-zero

CarefulPlan1

Hmm. Back to the past? Leela (2600} vs Rybka 3 (3100).