Lc0 51010

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
jp
Posts: 758
Joined: Mon Apr 23, 2018 5:54 am

Re: Lc0 51010

Post by jp » Fri Apr 05, 2019 10:01 am

mclane wrote:
Thu Apr 04, 2019 7:35 pm
And it’s also easier to concentrate on search and depths instead of understanding,
So what is your definition of understanding?

User avatar
mclane
Posts: 16086
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Fri Apr 05, 2019 10:52 am

Analyzing the position without search.
How do you think Mephisto III was capable to play chess if it only computed 3-5 NPS if there would not have been enough information about pins and threads and plans stored in the evaluation function ?
Somehow the program had to decide which important moves to compute if you do only these few NPS .
If you compute millions of NPS it's not so problematical.
You must decide which moves to follow and which branches to ignore because it is not interfering with the main idea.
Understanding means the program knows chess vocabulary. And knows how to use and handle the patterns on the board. E.g. how do you stop an isolani. Which passed pawns are dangerous, which not.
What is a good bishop. How to create a mate net. Can this blocked center be opened? Stupid programs exchange important own pieces against unimportant pieces of the opponent. Or do not connect more pieces into the attack.
This can all be done without search.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

jp
Posts: 758
Joined: Mon Apr 23, 2018 5:54 am

Re: Lc0 51010

Post by jp » Fri Apr 05, 2019 12:31 pm

That's a poor definition. It's not just about nps. It's about all of the brute-force computing power used.
It's computing rather than "understanding" whether or not that computing is searching nodes or doing something else.

I don't think Mephisto III would beat a modern program restricted to the same nps.

Fulvio
Posts: 146
Joined: Fri Aug 12, 2016 6:43 pm

Re: Lc0 51010

Post by Fulvio » Fri Apr 05, 2019 1:00 pm

jp wrote:
Fri Apr 05, 2019 12:31 pm
I don't think Mephisto III would beat a modern program restricted to the same nps.
You picked the wrong word.
You believe that a modern program plays decently even with restricted NPS.
If you thought about it....
Let's see how Stockfish behave in the previous position, 200 seconds per 5NPS = 1000 nodes:

Code: Select all

Stockfish 10 64 by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
uci
id name Stockfish 10 64
id author T. Romstad, M. Costalba, J. Kiiski, G. Linscott

option name Debug Log File type string default
option name Contempt type spin default 24 min -100 max 100
option name Analysis Contempt type combo default Both var Off var White var Black var Both
option name Threads type spin default 1 min 1 max 512
option name Hash type spin default 16 min 1 max 131072
option name Clear Hash type button
option name Ponder type check default false
option name MultiPV type spin default 1 min 1 max 500
option name Skill Level type spin default 20 min 0 max 20
option name Move Overhead type spin default 30 min 0 max 5000
option name Minimum Thinking Time type spin default 20 min 0 max 5000
option name Slow Mover type spin default 84 min 10 max 1000
option name nodestime type spin default 0 min 0 max 10000
option name UCI_Chess960 type check default false
option name UCI_AnalyseMode type check default false
option name SyzygyPath type string default <empty>
option name SyzygyProbeDepth type spin default 1 min 1 max 100
option name Syzygy50MoveRule type check default true
option name SyzygyProbeLimit type spin default 7 min 0 max 7
uciok
position fen 4r2k/q5pp/8/3Q2N1/8/8/5PPP/6K1 w - - 3 1
go nodes 1000
info depth 1 seldepth 2 multipv 1 score cp 165 nodes 90 nps 18000 tbhits 0 time 5 pv g5f7 h8g8 f7d6 g8f8 d6e8 f8e8
info depth 2 seldepth 4 multipv 1 score cp 218 nodes 147 nps 29400 tbhits 0 time 5 pv g5f7 h8g8 f7d6 g8f8 d5f3 f8e7
info depth 3 seldepth 6 multipv 1 score cp 501 nodes 230 nps 38333 tbhits 0 time 6 pv g5f7 h8g8 f7d6 g8f8 d5f3 f8g8 d6e8
info depth 4 seldepth 7 multipv 1 score cp 0 nodes 460 nps 65714 tbhits 0 time 7 pv g5f7 h8g8 f7g5 g8h8
info depth 5 seldepth 5 multipv 1 score cp 0 nodes 624 nps 78000 tbhits 0 time 8 pv g5f7 h8g8 f7g5 g8h8
info depth 6 seldepth 11 multipv 1 score cp 0 nodes 939 nps 104333 tbhits 0 time 9 pv g5f7 h8g8 f7d8 g8f8 d5f5 f8g8
info depth 7 seldepth 11 multipv 1 score cp 0 nodes 1005 nps 111666 tbhits 0 time 9 pv g5f7 h8g8 f7d8 g8f8 d5f5 f8g8
bestmove g5f7 ponder h8g8

Werewolf
Posts: 1192
Joined: Thu Sep 18, 2008 8:24 pm

Re: Lc0 51010

Post by Werewolf » Fri Apr 05, 2019 1:10 pm

SF > Mephisto at same nps.

I suspect it's less clear at same depth though.

User avatar
mclane
Posts: 16086
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Fri Apr 05, 2019 1:26 pm

Exactly. Stockfish is artificial stupidity. It concentrates on search instead of knowledge. Take away the millions and millions of NPS and you see the artificial stupidity.

It is more a pocket calculator with enormous power then an entity that can play chess.

Of course , as we can see, this stupidity combined with millions of NPS is enough to reach that high Elo.
Disappointing, isn't it ?

I thought there must be a more clever, a more artificial intelligent way to do so. Elo alone is not satisfying IMO.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

Uri Blass
Posts: 8558
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Lc0 51010

Post by Uri Blass » Fri Apr 05, 2019 3:01 pm

Werewolf wrote:
Fri Apr 05, 2019 1:10 pm
SF > Mephisto at same nps.

I suspect it's less clear at same depth though.
I am sure that you are right for big number of nodes.
I am less sure if you are right if both programs get small number of nodes.

I think that the main advantage of stockfish it that it has more intelligent search so it earns more elo from doubling the number of nodes when you start at the same playing strength.

Uri Blass
Posts: 8558
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Lc0 51010

Post by Uri Blass » Fri Apr 05, 2019 3:06 pm

mclane wrote:
Fri Apr 05, 2019 1:26 pm
Exactly. Stockfish is artificial stupidity. It concentrates on search instead of knowledge. Take away the millions and millions of NPS and you see the artificial stupidity.

It is more a pocket calculator with enormous power then an entity that can play chess.

Of course , as we can see, this stupidity combined with millions of NPS is enough to reach that high Elo.
Disappointing, isn't it ?

I thought there must be a more clever, a more artificial intelligent way to do so. Elo alone is not satisfying IMO.
I think that search is knowledge.
It is clear that Stockfish is more intelligent when it search millions of nodes relative to Mephisto and not because stockfish search more nodes.

If you give stockfish exactly the same size of hash tables like Mephisto and exactly the same number of million nodes per move then I expect stockfish to win easily.

Robert Pope
Posts: 510
Joined: Sat Mar 25, 2006 7:27 pm

Re: Lc0 51010

Post by Robert Pope » Fri Apr 05, 2019 3:31 pm

Uri Blass wrote:
Fri Apr 05, 2019 3:01 pm
Werewolf wrote:
Fri Apr 05, 2019 1:10 pm
SF > Mephisto at same nps.

I suspect it's less clear at same depth though.
I am sure that you are right for big number of nodes.
I am less sure if you are right if both programs get small number of nodes.

I think that the main advantage of stockfish it that it has more intelligent search so it earns more elo from doubling the number of nodes when you start at the same playing strength.
I am sure he is right for a small number of nodes, as well. The evaluation of Stockfish is an order of magnitude more nuanced and better turned than what Mephisto has.

User avatar
mclane
Posts: 16086
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Fri Apr 05, 2019 4:45 pm

How can I play a game of chess with Stockfish doing only 1000 Nodes per move.
Let’s say Mephisto III S computes 3-5 NPS that is 900 Nodes per 3 minute computation.
How can I run Stockfish 10 with doing only 1000 Nodes and I can run a match between the 2 chess programs.


When Mephisto III S came out in 1984 (8 bit Version came earlier and did only 1-3 NPS) it was a new approach.
But it was good enough to win a championship title against opponents who computed 1500 NPS.
So it was able to handle opponents of A strategy programs or AB programs doing 500 times (!!!) more NPS.


What has changed to today is that hardware development made progress, enormous progress.
In the old days opponents had 6502 with 5 MHz 8 bit. No hash.
Today Stockfish is doing millions of NPS.
That is a factor of 2000 if we use 3.000.000 NPS .
But as you know 3.000.000 is a slow cpu.

So a factor of 2000 to the situation in 1984.


That’s much.

I think the idea to use knowledge was realised very early in computerchess.
Do we really need these high amounts of computation for playing chess ?
Do we need these huge search trees ??
The search tree of Mephisto III could have been printed out in a piece of paper.
It must have been easy for the programmers to follow why their program played this or that with such a small tree.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

Post Reply