greetings. I expect that soon I will have something to show for recent efforts on my engine. In this phase I plan to have a basic "negamax recursion" algorithm with alpha beta pruning and quiescense. There will also be some primitive methods to eliminate some move transpositions. I've had lots of good input from people here, so thanks for that.
I plan to use pastebin.com to make source code available.
I've looked over the page on engine testing at the chessprogramming wiki, but I could use some suggestions on what type of output to generate in order to make it easy for interested parties here on TalkChess to evaluate the correctness of the algorithms, given a rather limited positional/material eval function.
Thanks.
What output is best for evaluating an engine ?
Moderators: hgm, Rebel, chrisw
-
- Posts: 814
- Joined: Sat May 09, 2009 4:51 pm
- Location: Toronto
-
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: What output is best for evaluating an engine ?
From Symbolic's finger notes:
Kibitzed analysis results are presented from Symbolic's point-of-view: [Score (decimal pawns)/Full width depth/CPU time used/Node count/Tablebase hit count] Predicted variation
So, something like:
[+0.423/3/0.055/32,654/0] 1 e4 e5 2 Nf3
Kibitzed analysis results are presented from Symbolic's point-of-view: [Score (decimal pawns)/Full width depth/CPU time used/Node count/Tablebase hit count] Predicted variation
So, something like:
[+0.423/3/0.055/32,654/0] 1 e4 e5 2 Nf3
-
- Posts: 892
- Joined: Sun Nov 19, 2006 9:16 pm
- Location: Russia
Re: What output is best for evaluating an engine ?
It is suggested to create a program playing complete chess and understands UCI or Winboard chess engine protocol.
Those protocols specify minimum required output.
Those protocols specify minimum required output.
-
- Posts: 814
- Joined: Sat May 09, 2009 4:51 pm
- Location: Toronto
Re: What output is best for evaluating an engine ?
Duly Noted, Steven and Aleks thanks
Implementing Winboard and/or UCI is on my wishlist for the future. Right now I have my own GUI and engine which communicate by a single method call, and also have a few shared libraries.
Still, I am somewhat familiar with the two protocols, so I'll look there also to get some ideas for suggested output.
regards.
Implementing Winboard and/or UCI is on my wishlist for the future. Right now I have my own GUI and engine which communicate by a single method call, and also have a few shared libraries.
Still, I am somewhat familiar with the two protocols, so I'll look there also to get some ideas for suggested output.
regards.
-
- Posts: 3562
- Joined: Thu Mar 09, 2006 3:54 am
- Location: San Jose, California
Re: What output is best for evaluating an engine ?
I wrote a very simple chess program that runs under Winboard and the only thing my program output was the move selected and Winboard was happy with that.Aleks Peshkov wrote:It is suggested to create a program playing complete chess and understands UCI or Winboard chess engine protocol.
Those protocols specify minimum required output.
Bill
-
- Posts: 1822
- Joined: Thu Mar 09, 2006 11:54 pm
- Location: The Netherlands
Re: What output is best for evaluating an engine ?
I feel you put a lot of faith in the 'community' to test your engine.Fguy64 wrote:greetings. I expect that soon I will have something to show for recent efforts on my engine. In this phase I plan to have a basic "negamax recursion" algorithm with alpha beta pruning and quiescense. There will also be some primitive methods to eliminate some move transpositions. I've had lots of good input from people here, so thanks for that.
I plan to use pastebin.com to make source code available.
I've looked over the page on engine testing at the chessprogramming wiki, but I could use some suggestions on what type of output to generate in order to make it easy for interested parties here on TalkChess to evaluate the correctness of the algorithms, given a rather limited positional/material eval function.
Thanks.
When i see new engines regurarly and inform its author, it's obvious to me usually that no one else tested it before other than the author.
Always go for an output you can readyourself very well.
Ignore Edwards type guys who like to overcomplicate things and put a stamp on it of their own
KISS is the way to go.
No hard feelings Edwards!
Happens a lot.
-
- Posts: 814
- Joined: Sat May 09, 2009 4:51 pm
- Location: Toronto
Re: What output is best for evaluating an engine ?
thanks, duly noted. I don't think I have too great expectations here, I have been coding and revising and testing on my own till the cows come home, for quite a long time, and I welcome the chance to bounce some ideas, and I just want to make it easy for people to bounce back. People here have been helpful, if they test, then great, if they don't well that's fine too, I still got my money's worth from this place.diep wrote:I feel you put a lot of faith in the 'community' to test your engine.Fguy64 wrote:greetings. I expect that soon I will have something to show for recent efforts on my engine. In this phase I plan to have a basic "negamax recursion" algorithm with alpha beta pruning and quiescense. There will also be some primitive methods to eliminate some move transpositions. I've had lots of good input from people here, so thanks for that.
I plan to use pastebin.com to make source code available.
I've looked over the page on engine testing at the chessprogramming wiki, but I could use some suggestions on what type of output to generate in order to make it easy for interested parties here on TalkChess to evaluate the correctness of the algorithms, given a rather limited positional/material eval function.
Thanks.
When i see new engines regurarly and inform its author, it's obvious to me usually that no one else tested it before other than the author.
Always go for an output you can readyourself very well.
Ignore Edwards type guys who like to overcomplicate things and put a stamp on it of their own
KISS is the way to go.
No hard feelings Edwards!
Happens a lot.
The chess part of this project is really kind of secondary, this project is mostly about learning to program in Java, and the chess makes it interesting.
regards.