Winboard Forum

Hi all,
Roce (which stands for Roman?s Own Chess Engine) knows finally how to mate. It took me quite some time to implement a somehow working alpha-beta search in my chess engine but know it almost looks as if it?s working.
By trying to create a search which doesn?t search any longer if a mate is found I created a strange bug which took me quite some time till I found it. Roce didn?t mate the opponent and instead did just about everything to not mate the opponent. Was really strange to look the engine winning the game and then always avoiding the winning move and instead even sacrifying pieces in order to not mate the opponent.

But the ?release? of this engine is also your chance to beat an engine that makes more than 500?000 kn/s in the middlegame (on an AMD Sempron 1800+) as there is only a pure alpha-beta search with no extensions at all and a simple material/square eval. I guess it?s one of the weakest engines around although it wins/draws against Cassandre regulary. As there is almost everything missing in Roce currently I hope to improve the playing strength a bit.

There is still a lot missing in Roce:
-Time management (calculates for 6s/move now only)
-Hash tables
-50-moves rules
-3-fold leads to draw
-some basic endgame knowldege (when to push the pawns in an endgame)
-winboard protocol (supports only UCI currently)

In case you want to try out Roce (keep in mind there is a reason that the current version # is 0.034) here is a link:

http://mypage.bluewin.ch/romanhartmann/roce03.zip

cheers
Roman

below is a game Roce vs Cassandre. It's quite a patzer game but then who am I to judge

[Event "Test25_07_05s"]
[Site "ATHLONXP2"]
[Date "2005.07.28"]
[Round "2"]
[White "Roce03"]
[Black "Cassandre"]
[Result "1-0"]
[ECO "A40"]
[Opening "Englund (Charlick) Gambit"]
[Time "22:08:23"]
[Variation "2.dxe5 Nc6"]
[TimeControl "0+5"]
[Termination "normal"]
[PlyCount "83"]
[WhiteType "program"]
[BlackType "program"]

1.d4 e5 {(e7e5 d4xe5 Nb8c6 Nb1c3 a7a5 Bc1e3 Nc6xe5) -0.10/6 3} 2.dxe5 Nc6
{(Nb8c6 Bc1f4 Bf8b4+ Nb1c3 Nc6xe5) -0.04/5 2} 3.Qd5 {+0.23/6 6} Qe7 {(Qd8e7 Ng1f3 Nc6b4
Qd5d1 Qe7xe5) -0.05/5 6} 4.Nf3 {+1.06/6 6} g6 {(g7g6 Bc1g5 Qe7b4+ Nb1c3 Nc6xe5)
-0.11/5 7} 5.Nc3 {+0.20/6 6} Bg7 {(Bf8g7 Bc1f4 Nc6xe5 Nf3xe5 Qe7xe5) -0.10/5 9} 6.Bg5
{+0.96/6 6} Qb4 {(Qe7b4 Ra1b1 Ke8f8 Bg5d2 Nc6xe5 Nf3xe5 Qb4xc3) -0.06/4 1} 7.Qb3
{+0.19/6 6} Qxb3 {(Qb4xb3 a2xb3 h7h6 Bg5e3 Nc6xe5) -0.06/4 1} 8.axb3 {+7.40/7 6} Nxe5
{(Nc6xe5 Nf3xe5 Bg7xe5 e2e4 d7d5) -0.10/5 3} 9.Nd5 {+1.08/7 6} Nxf3+ {(Ne5xf3+ e2xf3
Bg7xb2 Nd5xc7+ Ke8f8 Nc7xa8 Bb2xa1) -0.04/5 2} 10.exf3 {+0.69/8 6} Bxb2 {(Bg7xb2
Nd5xc7+ Ke8f8 Nc7xa8 Bb2xa1) -0.04/5 1} 11.Nxc7+ {+1.44/7 6} Kf8 {(Ke8f8 Ra1a2 Bb2c3+
Ke1d1 Ra8b8 Ra2xa7 d7d5) -0.93/7 1} 12.Ra2 {+5.96/7 6} Bc3+ {(Bb2c3+ Bg5d2 Bc3xd2+
Ke1xd2 Ra8b8 Ra2xa7 d7d5) -0.88/7 5} 13.Ke2 {+4.90/8 6} Rb8 {(Ra8b8 Bg5f4 Bc3g7 Nc7e6+
d7xe6 Bf4xb8) -1.77/6 2} 14.Bf4 {(Nc7d5 Bc3g7 Bg5f4 h7h5 f7h6) +2.20/7 6} Bg7 {(Bc3g7
Nc7d5 Rb8a8 Nd5b6 g6g5 Nb6xa8 g5xf4) -1.74/6 12} 15.Bd6+ {+5.98/7 6} Ne7 {(Ng8e7
Nc7d5 Bg7d4 Nd5xe7 Rb8a8 Ne7xc8+ Ra8xc8) -3.00/6 1} 16.Nd5 {+6.29/7 6} Ke8 {(Kf8e8
Bd6xe7 Bg7d4 Be7f6 Bd4xf6 Nd5xf6+) -3.08/6 3} 17.Nxe7 {+6.04/7 6} Ra8 {(Rb8a8 Ne7d5
Ke8d8 Ke2d3 Rh8e8 c2c4) -3.18/6 2} 18.Nd5 {+7.76/7 6} Kd8 {(Ke8d8 Bd6e7+ Kd8e8 Nd5c7+
Ke8xe7 Nc7xa8) -4.88/6 20} 19.Bc7+ {+7.87/8 6} Ke8 {(Kd8e8 Nd5b6 Ra8b8 Ra2xa7 Bg7d4
Bc7xb8 Bd4xb6) -5.93/6 1} 20.Nb6 {+11.61/7 6} Rb8 {(Ra8b8 Ra2xa7 Bg7d4 Bc7xb8 Bd4xb6
Ra7a1 Bb6xf2) -4.93/6 2} 21.Rxa7 {+4.64/7 6} Bd4 {(Bg7d4 Bc7xb8 Bd4xb6 Ra7a2 d7d5
Bb8f4 Bb6xf2) -4.97/6 1} 22.Ra4 {+6.91/7 6} Bxb6 {(Bd4xb6 Bc7xb8 d7d5 Bb8e5 Rh8f8
Be5f4 Bb6xf2) -4.97/7 2} 23.Re4+ {(Bc7xb8 d7d5 Bb6c7 Bc8h3 g6g5) +6.20/7 6} Kf8
{(Ke8f8 Bc7xb8 d7d5 Bb8d6+ Kf8g7 Re4b4 Rh8e8+ Ke2d1 Bb6xf2) -5.05/8 18} 24.Bxb8
{+3.11/8 6} d5 {(d7d5 Bb8d6+ Kf8g7 Re4b4 Rh8e8+ Ke2d1 Bb6xf2) -5.05/6 4} 25.Rb4 {+0.16/8
6} Bc5 {(Bb6c5 Rb4b5 b7b6 Bb8e5 Rh8g8 Be5f4 Bc5xf2) -4.88/6 1} 26.Rb5 {+6.92/7 6}
b6 {(b7b6 Rb5xc5 b6xc5 Bb8d6+ Kf8g8 Bd6xc5 Bc8f5) -4.75/7 4} 27.Rxc5 {+7.11/7 6}
bxc5 {(b6xc5 Ke2d1 Bc8f5 Bb8d6+ Kf8g8 Bd6xc5 Bf5xc2+) -3.87/7 2} 28.c4 {+4.04/8 6}
Ba6 {(Bc8a6 Bb8d6+ Kf8g8 Bd6xc5 f7f5 Ke2d1 Ba6xc4) -3.74/6 1} 29.Bd6+ {+5.50/8 6}
Kg7 {(Kf8g7 Ke2d1 d5xc4 Bd6xc5 Rh8d8+ Kd1c1 Rd8d5 Bc5e3 c4xb3) -3.12/8 2} 30.Bxc5
{+2.16/8 6} Rb8 {(Rh8b8 b3b4 Ba6xc4+ Ke2d2 Bc4xf1 Rh1xf1 Rb8xb4) -3.03/7 2} 31.Bd4+
{+3.23/8 6} Kg8 {(Kg7g8 Ke2d1 Rb8xb3 Kd1c1 Ba6xc4 Bf1xc4 d5xc4 f3f4) -2.92/8 3} 32.Kd2
{+5.78/7 6} Rxb3 {(Rb8xb3 Kd2c1 d5xc4 f3f4 f7f5 Bf1e2 Ba6b7) -2.86/7 1} 33.cxd5 {+5.90/7
6} Rb1 {(Rb3b1 d5d6 Ba6c8 g2g4 Rb1b4 Bd4e3 Rb4b1 d6d7 Rb1xf1) -2.27/8 4} 34.Bxa6
{+6.09/7 6} Rxh1 {(Rb1xh1 d5d6 Kg8f8 Bd4g7+ Kf8e8 Ba6b5+ Ke8d8 Bb5d3 Rh1xh2) -2.34/8 10}
35.d6 {+2.48/8 6} Kf8 {(Kg8f8 Bd4g7+ Kf8e8 Ba6b5+ Ke8d8 Bb5d3 Rh1xh2) -2.34/6 2}
36.Bf6 {+3.33/8 6} Rb1 {(Rh1b1 d6d7 Rb1b2+ Bf6xb2 Kf8e7 Ba6d3 Ke7xd7) -7.23/6 2} 37.d7
{+6.58/8 6} Rb8 {(Rb1b8 Ba6c8 Rb8b2+ Kd2c1 Rb2c2+ Kc1xc2) -8.07/6 5} 38.Bc8 {+2.88/8 6}
Rb2+ {(Rb8b2+ Bf6xb2 Kf8e7 Bb2f6+ Ke7xf6 d7d8Q+) -13.99/6 3} 39.Kd3 {+19.13/9 6}
Rb3+ {(Rb2b3+ Kd3c2 Rb3c3+ Kc2xc3 Kf8g8 d7d8Q+) -16.81/6 3} 40.Kc4 {+3.22/8 6} Re3
{(Rb3e3 f2xe3 g6g5 d7d8Q+ g5g4 Qd8xf8) -M500/6 2} 41.fxe3 {+8.35/8 6} Kg8 {(Kf8g8 g2g3
g6g5 d7d8Q+ g5g4 f3xg4+ h7h6 Qd8xg8) -M500/8 5} 42.d8=Q# {+13.43/9 6} {Mate} 1-0

Hi Roman,

Thanks a lot for Roce! If I only had a Windows computer, I would have given it a try. It's nice to have a few engines agaist which even we lowly patzers can score a lucky win once in a while, even if it is just for a short time. If a few months, it will probably crush us all, if you keep working on it.

You have finished the most difficult and boring part of the work now. I think you will find that chess programming is much more fun when you have a complete, working engine, and can enjoy watching its rapid progress.

There is still a lot missing in Roce:
-Time management (calculates for 6s/move now only)
-Hash tables
-50-moves rules
-3-fold leads to draw
-some basic endgame knowldege (when to push the pawns in an endgame)
-winboard protocol (supports only UCI currently)

I would recommend you to implement time management, 50 move draws and 3-fold repetitions first, because it would enable your engine to participate in the many nice tournaments for amateur engines. It is great fun to watch your engine slowly crawl towards the top divisions.

Good luck with Roce!

Tord

Hi Tord,
thanks for your kind words. I'm glad my program does finally something more than just crash. The next step is certainly to implement the time management (in fact I have already some code but it doesn't work properly with incremental time controls). And it's def. more fun now to play around with some working parts now than to spend endless hours looking for bugs in the move generator (and I had a lot of bugs there).
As soon as Roce is lacking less of the features I listed, I will create a Linux version as well.

cheers
Roman

I added time management to Roce in the meantime and also let it play some blitz games against some other low rated engines. According to this games I would estimate its playing strength to be in the 1300-1400 range.

cheers
Roman

Hi Roman,

I read your README, and it sounds as if you don't have qsearch implemented. If you haven't, you should probably add something like this to avoid simply losing material very early. Also I assume you will need any kind of move ordering improvement like killer moves or penalizing losing captures. Then your evaluation will need to know a bit about king safety, e.g. avoid running with the pawns guarding the king or something similar.

Roman Hartmann wrote:more than 500?000 kn/s

I assume 500 kn/s, right? Otherwise I would really like to see the code needing only 5 CPU cycles per node.

Sven

Sven Sch?le wrote:Hi Roman,

I read your README, and it sounds as if you don't have qsearch implemented. If you haven't, you should probably add something like this to avoid simply losing material very early. Also I assume you will need any kind of move ordering improvement like killer moves or penalizing losing captures. Then your evaluation will need to know a bit about king safety, e.g. avoid running with the pawns guarding the king or something similar.

Roman Hartmann wrote:more than 500?000 kn/s

I assume 500 kn/s, right? Otherwise I would really like to see the code needing only 5 CPU cycles per node.

Sven

Hi Sven,
you're right it's 'only' 500 kn/s. And I also don't have a QS yet although it should be relatively easy to implement. I also don't have killer moves yet and king safety is nothing that my engine is aware of. Currently my alpha-beta search is also called with a fully open window all the time as I have some search instabilities otherwise. But only adding QS should help to improve the play quite a bit as the evaluation is really unreliable at the moment and the engine definitely suffers from the horizon effect.

Roman