stockfish 10 vs. Mephisto III S Glasgow

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

Moderators: hgm, Rebel, chrisw

Madeleine Birchfield
Posts: 512
Joined: Tue Sep 29, 2020 4:29 pm
Location: Dublin, Ireland
Full name: Madeleine Birchfield

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by Madeleine Birchfield »

JohnW wrote: Tue Nov 30, 2021 8:09 pm Just like the analogy of the GPS navigation system mentioned earlier, it's not really planning a route from point A to point B. It identifies where you are now and where you want to go then triangulates. Then with that information it looks up the route in a database. Sure it may appear to be planning but it's really not.
In computer chess, it would be an evaluation function rather than a database, and a search algorithm rather than triangulation. After all, the space of all possible games is a graph, which is also the case with navigation systems and stuff like Google maps as well where the space of all locations is modeled as a graph.
User avatar
mhull
Posts: 13447
Joined: Wed Mar 08, 2006 9:02 pm
Location: Dallas, Texas
Full name: Matthew Hull

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mhull »

It seems to boil down to a human aesthetic. Do we marvel at computer games like we do the games from Morphy or Capablanca?

I think the answer has always been "yes, sometimes". Even TC must admit this as many of his own dedicated CC reviews testify. Back in the day the David Kittinger programs were admired for their natural game play. The Novag Super Constellation and Super Forte C come to mind in this regard -- the latter garnering praise, even from international masters.

Recently, the Alpha Zero project impressed top players with it's seemingly intuitive understanding of chess strategy, breathing new life into the game and delighting chess enthusiast with "beautiful games" in contrast to previous trend of incomprehensible contests between machines playing arcane and "disgusting engine lines".

So, they aren't all "bloody iron monsters".
Matthew Hull
dkappe
Posts: 1632
Joined: Tue Aug 21, 2018 7:52 pm
Full name: Dietrich Kappe

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by dkappe »

Madeleine Birchfield wrote: Tue Nov 30, 2021 5:59 pm How is reinforcement training neural networks, like what Tomasz Sobczyk is doing right now for Stockfish and what Dietrich Kappe is doing for Komodo Dragon, not teaching an a chess engine how to plan? Or if reinforcement training isn't giving a chess engine knowledge, then why is Leela, which Thorsten Czub likes to claim is different from Stockfish, any different from Stockfish and not just like another pocket calculator?
The stockfish team isn’t doing Reinforcement Learning (RL). They are using leela data (which is actually RL data) with some additional self generated correction data. That’s not a criticism, just an observation. They are able to do quite a bit with filtering the data and tweaking the training software.

Dragon, though, is doing reinforcement learning, where each subsequent net generates new data and the old data drops out the back of the training window.
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".
User avatar
mhull
Posts: 13447
Joined: Wed Mar 08, 2006 9:02 pm
Location: Dallas, Texas
Full name: Matthew Hull

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mhull »

mclane wrote: Mon Nov 29, 2021 9:10 pm Back to the topic.
A plan is not a main line

Stockfish main line and score shows it has no plan.
One might observe that a plan is a human way of coping with his limited ability to search the tree of possibilities. Computers search rather than plan because its simpler and more natural to implement on a computer than to implement planning on a computer. And searching just works. If something works, humans are reluctant to fix it.

Not that it hasn't been tried,though:
https://www.chessprogramming.org/Paradise

Steven Edwards tried to build upon the Paradise idea, to create a program that would make plans around key features detected in a position. Somewhere along the way, the effort slowed and then stopped. AFAIK, he never talked much about why the project stalled. He continued to refine the A/B searcher function of his program Symbolic (which could play whole games in itself) but the higher level planning functions were never finished.
Matthew Hull
User avatar
mclane
Posts: 18764
Joined: Thu Mar 09, 2006 6:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mclane »

What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
User avatar
mhull
Posts: 13447
Joined: Wed Mar 08, 2006 9:02 pm
Location: Dallas, Texas
Full name: Matthew Hull

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mhull »

mclane wrote: Tue Dec 07, 2021 8:31 am This here ?!
http://www.ai.sri.com/pubs/files/465.pdf
Affirmative. Steven Edwards' Symbolic was intended to take the Paradise concept to the next level.
Matthew Hull
User avatar
mclane
Posts: 18764
Joined: Thu Mar 09, 2006 6:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mclane »

Did it play games ?! Or was it unfinished?
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
User avatar
mhull
Posts: 13447
Joined: Wed Mar 08, 2006 9:02 pm
Location: Dallas, Texas
Full name: Matthew Hull

Re: stockfish 10 vs. Mephisto III S Glasgow

Post by mhull »

mclane wrote: Wed Dec 08, 2021 9:13 pm Did it play games ?! Or was it unfinished?
Paradise was unfinished. It could solve test positions but couldn't play a full game of chess.

Symbolic could play full games using the toolkit A/B searcher. But it could not play a game using the plan-production functions which was the goal. In that sense, it too was unfinished.
Matthew Hull