Symbolic vs FairyMax: match results
Another 100 game match using a five minute sudden death time control; the score: 72-17-11 (+215 elo)
The games: https://dl.dropboxusercontent.com/u/316 ... 0-1.pgn.gz
Symbolic vs FairyMax: match results
Moderators: hgm, Rebel, chrisw
-
- Posts: 27817
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Symbolic vs FairyMax: match results
So this also starts to approach the regime where the opponent is too weak to usefully compare performance of the engines-under-test. The match of 100 5-min games between King Slayer/Simple and Fairy-Max I ran tonight ended at 86% (78+, 16=, 6-), which is similar or slightly better than Symbolic's result against Fairy-Max.
So now that Symbolic decisively beats an engine without move ordering or knowledge, you could try it against King Slayer to see how it does against an engine with reasonable move ordering (MVV/LVA and killers), but still without knowledge (other than piece values through PST).
So now that Symbolic decisively beats an engine without move ordering or knowledge, you could try it against King Slayer to see how it does against an engine with reasonable move ordering (MVV/LVA and killers), but still without knowledge (other than piece values through PST).
-
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: Symbolic vs FairyMax: match results
From http://www.3dkingdoms.com/chess/elo.htmhgm wrote:The match of 100 5-min games between King Slayer/Simple and Fairy-Max I ran tonight ended at 86% (78+, 16=, 6-), which is similar or slightly better than Symbolic's result against Fairy-Max.
78 win / 6 lose / 16 draw = +315 elo
-
- Posts: 193
- Joined: Wed Mar 11, 2015 3:34 am
- Location: United States
Re: Symbolic vs FairyMax: match results
I've been performing similar tournaments while testing my latest engine: Clunkhgm wrote:So this also starts to approach the regime where the opponent is too weak to usefully compare performance of the engines-under-test. The match of 100 5-min games between King Slayer/Simple and Fairy-Max I ran tonight ended at 86% (78+, 16=, 6-), which is similar or slightly better than Symbolic's result against Fairy-Max.
So now that Symbolic decisively beats an engine without move ordering or knowledge, you could try it against King Slayer to see how it does against an engine with reasonable move ordering (MVV/LVA and killers), but still without knowledge (other than piece values through PST).
Where can I get a copy of King Slayer?
-
- Posts: 27817
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
-
- Posts: 193
- Joined: Wed Mar 11, 2015 3:34 am
- Location: United States
Re: Symbolic vs FairyMax: match results
Thanks H.G. That's some impressive coding! Very small yet doing quite a bit. And pretty fast too.
Can't wait to see how Clunk fairs against it. Will King Slayer be playing in your next blitz tourney? I'll be registering Clunk if I can secure the time and a good internet connection.
-
- Posts: 27817
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Symbolic vs FairyMax: match results
Yes, King Slayer will participate, and in fact already did so on the two previous occasions. (First time under the handle 'Simple'.)
Note that King Slayer really is an unfinished project; most of the code in there associated with the evaluation (Pawn structure, King safety, patterns, end-game knowledge, mobility) is still switched off by default, and might not work correctly when switched on. It was not tested or debugged, and is sometimes not even finished, and most certainly not tuned.
I am still wrestling with the move ordering; the current version uses strict MVV/LVA, but I had envisaged a quirky method for postponing suspect HxL captures, by only 'conditionally' searching them, but aborting such a search as soon as it is clear that the victim was protected, or a counter threat against a more valuable piece exists, and then rescheduling the move until after the good captures for an unconditional search. This still doesn't work correctly.
Note that King Slayer really is an unfinished project; most of the code in there associated with the evaluation (Pawn structure, King safety, patterns, end-game knowledge, mobility) is still switched off by default, and might not work correctly when switched on. It was not tested or debugged, and is sometimes not even finished, and most certainly not tuned.
I am still wrestling with the move ordering; the current version uses strict MVV/LVA, but I had envisaged a quirky method for postponing suspect HxL captures, by only 'conditionally' searching them, but aborting such a search as soon as it is clear that the victim was protected, or a counter threat against a more valuable piece exists, and then rescheduling the move until after the good captures for an unconditional search. This still doesn't work correctly.