Hawkeye 1.01

[MikeB]

Still working on this derivative of Gull.

Thought this position might be of interest to some (see prior Gull posts):

http://talkchess.com/forum/viewtopic.ph ... t&start=60

[d]3k4/8/8/8/8/3N4/2BKR3/8 w - - 0 1

[pgn][Event "Computer Chess Game"]
[Site "Mac-Pro.local"]
[Date "2016.07.05"]
[Round "-"]
[White "Hawkeye 1.01 OSx64"]
[Black "Komodo 10 64-bit"]
[Result "1-0"]
[TimeControl "9+0"]
[FEN "3k4/8/8/8/8/3N4/2BKR3/8 w - - 0 1"]
[SetUp "1"]
1. Re6 {+12.75/63} Kc7 {-1000.08/18 0.1} 2. Ne5 {+1000.08/16 0.1} Kb7
{-1000.07/17 0.1} 3. Ke3 {+1000.07/17 0.2} Kc7 {-1000.06/18 0.1} 4. Kd4
{+1000.06/20 0.4} Kb7 {-1000.05/20 0.1} 5. Re7+ {+1000.05/22 0.4} Kb6
{-1000.04/21 0.1} 6. Kc4 {+1000.04/29 0.3} Ka5 {-1000.03/85 0.1} 7. Kc5
{+1000.03/40 0.3} Ka6 {-1000.02/5 0.1} 8. Bb3 {+1000.02/63 0.1} Ka5
{-1000.01/5 0.1} 9. Ra7# {+1000.02/63 4}
{White mates} 1-0[/pgn]

As well at hitting the mate score on this position in 37 seconds:
[d]8/1Q5p/3B1k2/8/8/1p3P1P/4q1PK/8 w - - 3 1

Code: Select all

info time 37114 depth 25 seldepth 62 score cp 30000 nps 26045726 nodes 966661078 tbhits 199268 pv f3f4 b3b2 b7b2 e2b2 d6e5 f6e6 e5b2 e6d6 h2g3 d6e6 g3h4 e6f5 g2g4 f5e4 f4f5 e4d5 h4g5 d5c4 f5f6 c4d3 f6f7 d3e4 f7f8q e4e3 f8f7 h7h5 g5h5 e3e2 g4g5 e2d1 g5g6

[pgn][Event "Computer Chess Game"]
[Site "Mac-Pro.local"]
[Date "2016.07.05"]
[Round "-"]
[White "Hawkeye 1.01 OSx64"]
[Black "Komodo 10 64-bit"]
[Result "1-0"]
[TimeControl "9+0"]
[FEN "8/1Q5p/3B1k2/8/8/1p3P1P/4q1PK/8 w - - 3 1"]
[SetUp "1"]
1. f4 {+300.00/23} b2 {-23.30/17 0.6} 2. Qxb2+ {+300.00/18 0.3} Qxb2
{-250.00/27 0.2} 3. Be5+ {+300.00/21 0.1} Ke6 {-250.00/28 0.1} 4. Bxb2
{+300.00/19 0.2} Kf7 {-250.00/27 0.1} 5. Bd4 {+300.00/24 0.6} Kg6
{-250.00/28 0.2} 6. g4 {+300.00/24 0.3} Kf7 {-250.00/26 0.1} 7. f5
{+300.00/24 0.7} Ke7 {-250.00/24 0.1} 8. f6+ {+300.00/21 0.3} Kf7
{-250.00/25 0.1} 9. g5 {+300.00/21 0.5} Kg6 {-250.00/27 0.1} 10. Bc5
{+300.00/24 0.3} Kf5 {-250.00/28 0.1} 11. g6 {+300.00/20 0.4} h5
{-250.00/30 0.1} 12. g7 {+300.00/20 0.3} Ke6 {-250.00/35 0.1} 13. f7
{+300.00/22 0.3} Kxf7 {-250.00/99 0.1} 14. Bd4 {+5.29/63 0.1} Kg8
{-1000.13/20 0.1} 15. Kg3 {+300.00/41 0.3} Kh7 {-1000.10/22 0.1} 16. g8=Q+
{+300.00/41 0.3} Kxg8 {-1000.22/21 0.1} 17. Kh4 {+0.00/63 0.1} Kf7
{-1000.18/26 0.1} 18. Kxh5 {+300.00/32 0.3} Ke6 {-1000.16/26 0.1} 19. h4
{+300.00/29 0.3} Kd5 {-1000.15/25 0.1} 20. Bf6 {+300.00/31 0.4} Ke4
{-1000.13/24 0.1} 21. Kg5 {+300.00/32 0.3} Kd3 {-1000.11/24 0.1} 22. h5
{+300.00/32 0.3} Kc2 {-1000.10/25 0.1} 23. h6 {+300.00/31 0.3} Kb3
{-1000.09/25 0.1} 24. h7 {+300.00/30 0.3} Kc4 {-1000.08/25 0.1} 25. h8=Q
{+300.00/28 0.2} Kb4 {-1000.06/25 0.1} 26. Qc8 {+1000.05/14 0.1} Kb5
{-1000.05/29 0.1} 27. Bd4 {+1000.05/18 0.4} Kb4 {-1000.05/29 0.1} 28. Qc2
{+1000.04/21 0.3} Kb5 {-1000.04/43 0.1} 29. Qc3 {+1000.04/22 0.4} Ka4
{-1000.03/99 0.1} 30. Qb2 {+1000.03/28 0.2} Ka5 {-1000.02/5 0.1} 31. Qb3
{+1000.02/37 0.5} Ka6 {-1000.01/5 0.1} 32. Qb6# {+1000.01/38 1.0}
{White mates} 1-0
[/pgn]

[MikeB]

The only issue with these "features" is that they appear to reduce playing strength measurably.

[Ozymandias]

The only issue with these "features" is that they appear to reduce playing strength measurably.

Wasn't it stronger in testing?

[MikeB]

The only issue with these "features" is that they appear to reduce playing strength measurably.

Wasn't it stronger in testing?

latest, played 5 sets of 180 games , each set consisted of two positions chosen at random from a pool of 2500 openings, played b&w at time control of 18 sec base with .3 sec inc

Code: Select all

4 rounds and 180 games completed...
time control: 1800+30
Date: 07/06/16 : 22:46:52
Rank Name                     Elo    +    - games score oppo. draws 
   1 Stockfish DD 64 SSE4.2  3324   76   76    36   69%  3186   39% 
   2 Mikhail 0.04 OSx 64     3284   75   75    36   64%  3191   44% 
   3 Hawkeye 1.01 OSx64      3276   75   75    36   61%  3192   44% 
   4 Critter 1.6a 64-bit     3259   76   76    36   58%  3193   39% 
   5 Texel 1.06a34 64-bit    3217   78   78    36   53%  3198   33% 
   6 Gull 3 x64 (syzygy)     3216   76   76    36   51%  3198   36% 
   7 Hawkeye 1.0 x64         3212   77   77    36   50%  3199   33% 
   8 Hakkapeliitta TCEC v2   3146   78   78    36   42%  3206   28% 
   9 Kōhai 1.0               3035   85   85    36   26%  3218   14% 
  10 Crafty-25.1.Syzy-0611d  3032   83   83    36   25%  3219   22% 

4 rounds and 180 games completed...
time control: 1800+30
Date: 07/06/16 : 23:05:59
Rank Name                     Elo    +    - games score oppo. draws 
   1 Hawkeye 1.01 OSx64      3326   74   74    36   69%  3186   33% 
   2 Hawkeye 1.0 x64         3285   71   71    36   64%  3191   44% 
   3 Gull 3 x64 (syzygy)     3250   69   69    36   58%  3194   56% 
   4 Stockfish DD 64 SSE4.2  3242   71   71    36   57%  3195   47% 
   5 Texel 1.06a34 64-bit    3205   74   74    36   50%  3199   22% 
   6 Mikhail 0.04 OSx 64     3204   70   70    36   51%  3200   53% 
   7 Critter 1.6a 64-bit     3171   71   71    36   44%  3203   44% 
   8 Kōhai 1.0               3166   76   76    36   44%  3204   28% 
   9 Hakkapeliitta TCEC v2   3113   70   70    36   36%  3210   44% 
  10 Crafty-25.1.Syzy-0611d  3039   76   76    36   25%  3218   28% 

4 rounds and 180 games completed...
time control: 1800+30
Date: 07/06/16 : 23:23:35
Rank Name                     Elo    +    - games score oppo. draws 
   1 Gull 3 x64 (syzygy)     3301   69   69    36   65%  3189   47% 
   2 Mikhail 0.04 OSx 64     3300   73   73    36   65%  3189   25% 
   3 Stockfish DD 64 SSE4.2  3294   68   68    36   65%  3190   53% 
   4 Hawkeye 1.0 x64         3283   68   68    36   64%  3191   50% 
   5 Hawkeye 1.01 OSx64      3273   69   69    36   61%  3192   44% 
   6 Critter 1.6a 64-bit     3179   68   68    36   46%  3202   53% 
   7 Crafty-25.1.Syzy-0611d  3116   70   70    36   36%  3209   39% 
   8 Kōhai 1.0               3110   72   72    36   36%  3210   33% 
   9 Hakkapeliitta TCEC v2   3073   73   73    36   31%  3214   28% 
  10 Texel 1.06a34 64-bit    3071   74   74    36   31%  3214   28% 

4 rounds and 180 games completed...
time control: 1800+30
Date: 07/06/16 : 23:43:05
Rank Name                     Elo    +    - games score oppo. draws 
   1 Hawkeye 1.01 OSx64      3337   75   75    36   71%  3185   36% 
   2 Hawkeye 1.0 x64         3327   73   73    36   69%  3186   50% 
   3 Critter 1.6a 64-bit     3245   72   72    36   56%  3195   44% 
   4 Gull 3 x64 (syzygy)     3232   75   75    36   54%  3196   36% 
   5 Mikhail 0.04 OSx 64     3226   75   75    36   54%  3197   31% 
   6 Stockfish DD 64 SSE4.2  3201   72   72    36   49%  3200   42% 
   7 Hakkapeliitta TCEC v2   3184   77   77    36   47%  3202   28% 
   8 Kōhai 1.0               3166   74   74    36   44%  3204   39% 
   9 Texel 1.06a34 64-bit    3136   76   76    36   40%  3207   31% 
  10 Crafty-25.1.Syzy-0611d  2945   90   90    36   15%  3228   14% 
[Mac-Pro:~/cluster.mfb] michaelbyrne% 

in the last round Hawkeye 1.01 OSx64 went 19-4-13 (71%)while Hawkeye 1.0 x64 went 16-2-18 (69%) - so between the two Hawkeyes only six losses between them in 72 games . One of win/losses was between themselves - so just 5 losses to the remaining 8 opponents in 64 games, a combined 34-5-25 (73%). Looks like a release will be ready soon!

Combining all 5 sets into one result would look like this:

Code: Select all

20 rounds and 900 games completed...
time control: 1800+30
Date: 07/06/16 : 23:43:05
ResultSet-EloRating>ratings
Rank Name                     Elo    +    - games score oppo. draws 
   1 Hawkeye 1.01 OSx64      3255   39   39   180   65%  3138   38% 
   2 Hawkeye 1.0 x64         3234   39   39   180   62%  3141   42% 
   3 Stockfish DD 64 SSE4.2  3224   38   38   180   61%  3142   46% 
   4 Gull 3 x64 (syzygy)     3216   38   38   180   59%  3143   42% 
   5 Mikhail 0.04 OSx 64     3190   39   39   180   56%  3146   36% 
   6 Critter 1.6a 64-bit     3161   38   38   180   51%  3149   45% 
   7 Texel 1.06a34 64-bit    3108   40   40   180   44%  3155   30% 
   8 Hakkapeliitta TCEC v2   3073   40   40   180   39%  3159   33% 
   9 Kōhai 1.0               3061   40   40   180   37%  3160   29% 
  10 Crafty-25.1.Syzy-0611d  2980   42   42   180   26%  3169   26% 

[Ozymandias]

The only issue with these "features" is that they appear to reduce playing strength measurably.

Combining all 5 sets into one result would look like this:

20 rounds and 900 games completed...
time control: 1800+30
Date: 07/06/16 : 23:43:05
ResultSet-EloRating>ratings

Code: Select all

Rank Name                     Elo    +    - games score oppo. draws 
   1 Hawkeye 1.01 OSx64      3255   39   39   180   65%  3138   38% 
   2 Hawkeye 1.0 x64         3234   39   39   180   62%  3141   42% 
   3 Stockfish DD 64 SSE4.2  3224   38   38   180   61%  3142   46% 
   4 Gull 3 x64 (syzygy)     3216   38   38   180   59%  3143   42% 

That's measurably stronger, not weaker.

[MikeB]

The "features" -e.g to have it play Re6 and not Re7 in the tower position had to be removed. Also not 100% convinced syzygy table base was correctly implemented , specifically as it deals with the 50 move rule, part of that stems from not understanding the original Gull source.

[velmarin]

To my understanding when is puts a position to a motor without played previous, in mode analisys the engine assumes that you have 50 moves by front. or perhaps it is not so?

If also in this position change that tower by a Queen,
My how human eyes bleed.
Seeing that move Queen e7.


[D]3k4/8/8/8/8/3N4/2BKQ3/8 w - - 0 1

[MikeB]

To my understanding when is puts a position to a motor without played previous, in mode analisys the engine assumes that you have 50 moves by front. or perhaps it is not so?

If also in this position change that tower by a Queen,
My how human eyes bleed.
Seeing that move Queen e7.


[D]3k4/8/8/8/8/3N4/2BKQ3/8 w - - 0 1

My reference was not the position. I was referring to the source. I failed to see a 50 move counter for pawn moves/captures-so not 100% certain how it would play correctly in cases if it's not tracking when a pawn move/capture was last played to account for the 50 move rule . Sure it might work fine in many cases if it just assumes it has 50 moves for every position - but in real world it may need to adjust a move so that the counter is reset. And maybe it's in the source, but the source is so obtuse I could not see it it.