Pigeon (new open source engine)

[StuartRiffle]

Hi all.

Pigeon is an original UCI engine which I've been puttering on for a while. If you're into that sort of thing, I invite you to have a look!

It's still a work in progress, but (as far as I know) fully functional, and I'd be very interested in any feedback/advice you may have.

Pigeon is open source (MIT license), and can be found on GitHub:
https://github.com/StuartRiffle/pigeon

There is an x64 binary of the latest build (v1.33) up there too.

It's not terribly strong yet. An older version ran as a bot on ICC/FICS for a few days, and seemed to play around 1900 elo against humans at blitz. (It can beat me, but that's not saying much). I have a lot of tuning to do.

Pigeon is a brute-force style negamax implementation, without much high-level chess knowledge, except for what is baked into the evaluation. The critical code is mostly branch-free, to allow for SIMD-ification (which is almost done) and future migration to GPU (which I haven't started).

It has no opening book, because pigeons can't read.

Anyway, it's available now for the curious, and if it gives you a good game of chess (or a bad one!), I'd love to hear.

Cheers,
-Stuart

[cdani]

Thanks! Good luck with it!

[Graham Banks]

All the best with Pigeon. I'll be following its progress with interest.

[Guenther]

Something seems wrong with this engine. It loses almost all of its matches heavily on my computer, even against 1700-odd opponents.
Simply by watching it play I can't tell what the problem may be.

I played it under Fritz 13 and Fritz 15.

There is now a version 1.35 available with a fixed pruning error.
No idea though if it fixes the problem you found.

(I also tested it out of curiosity and found it often blundered a piece
in the first 10 moves after start pgn ended. This was against Cecir 3.5,
which I hoped to be at similar strength. These were only 2/40 or 3/40 games though)


https://github.com/StuartRiffle/pigeon/releases

[StuartRiffle]

I'm so sorry about that! It really was very broken, as you saw.

I have verified that the latest release (1.36) plays complete, legal games of chess again. It's still not very strong, but it should not go wacky on you.

The dev build is also now up on ICC from time to time, playing blitz as "Pigeon".

[Guenther]

I'm so sorry about that! It really was very broken, as you saw. :(

I have verified that the latest release (1.36) plays complete, legal games of chess again. It's still not very strong, but it should not go wacky on you.

The dev build is also now up on ICC from time to time, playing blitz as "Pigeon".

Conditions:

Code: Select all

Win 7-64 Ultimate SP1
Q8200 4*2.33Ghz 4GB DDR2 Ram
----------------------------------------
Winboard 4.80b
Hash 256 MB
TC 240/40
Ponder Off
All debugs saved
----------------------------------------
GS70 start positions

Ordo rating after 74/140 games
(roughly adjusted to GURL = Grand Unified Rating List):

Code: Select all

   # PLAYER           :  RATING  POINTS  PLAYED   (%)
   1 Zotron 4.4.6     :  1682.2    54.5      74  73.6
   2 Pigeon_136-64    :  1497.8    19.5      74  26.4

White advantage = 55.69
Draw rate (equal opponents) = 50.00 %

Code: Select all

Cross table(after 76/140 games, sorted by score percentage, Buchholz, SB

                              Zotron 4.4.6  Pigeon_136-64
                              ############# 1=11111011011
                              ############# 1=0110101111=
 1. Zotron 4.4.6              ############# =1=11=01110==
                              ############# 1=1=111111110
                              ############# 110===1111===
                              ############# 11=01==1111     74%  56.5 (1482.0, 1101.8)

                              0=00000100100 #############
                              0=1001010000= #############
 2. Pigeon_136-64             =0=00=10001== #############
                              0=0=000000001 #############
                              001===0000=== #############
                              00=10==0000   #############   26%  19.5 (4294.0, 1101.8)

More later. There are still some major quirks according to the PGN file, which should be sorted out.
I hope you will not become guilty of having me infected, with that damned computerchess virus, again?!?

Guenther

[StuartRiffle]

Cool! Thank you Guenther.

Would you post or PM me the PGN? I'm interested to see the (many) bad decisions it made...

[Guenther]

Cool! Thank you Guenther.

Would you post or PM me the PGN? I'm interested to see the (many) bad decisions it made...

I had some time to work again on my old RWBC pages.
You can find some tables/results/comments and the PGN files there.
There were 3 matches with 140 games each. The last match already
included 1.37.
It is not yet ready commented. As I did in the past there are general
comments about found bugs/limitations/etc. (only the obvious things to me),
because it is very time consuming to check all games intensively.

Either you go directly to the new page below, or via the RWBC index
=> Other tourneys => GS70 matches

http://rwbc-chess.de/GS70matches.htm

Code: Select all

Ordo Ratings after T1-T3 (Zotron 4.46 fixed to around 1682)
# PLAYER           :  RATING  ERROR  POINTS  PLAYED   (%)
1 Zotron 4.4.6     :  1682.4   33.6   182.5     280    65
2 Pigeon_137-64    :  1603.4   47.9    54.5     140    39
3 Matilde_261      :  1582.5   47.1    78.5     140    56
4 Pigeon_136-64    :  1539.7   32.2   104.5     280    37

White advantage = 8.98 +/- 14.62
Draw rate (equal opponents) = 31.70 % +/- 2.52

Guenther

P.S. It was an awful experience looking through all my old html/css code
from 2006 - it was not easy to resist some big overhauling ;-)