Scorpio & egbb issue (OSX)

[Daniel Shawul]

Pruning probably make it much harder. I try to turn off most of them when in EGBB but there are some such as Null move that I do not, and probably play a role in KBNK. Scorpio _without_ EGBBs failed to give a mate with depth=8 search for instance.
Edit Never mind, I forgot that I have removed the KBNK table from Scorpio's eval.

-Game removed-

[syzygy]

For KBNK mating you only need one table and nothing else. Infact mobility and other irrelevant evals could elongate finiding the mate. The dll provides you just that so it CAN'T be any worse.

It can certainly be worse if the engine uses bitbases within the tree instead of in the leaves. You're not going to search the same tree and the effective depth reached is likely lower than what the engine would otherwise have reached. (And if you probe at the leaves instead of within the tree, you're going to lose most of the advantage of bitbases.)

It can also very easily be worse if there is any possibility of evaluation scores being compared to heuristic scores returned by your dll.

[Daniel Shawul]

For KBNK mating you only need one table and nothing else. Infact mobility and other irrelevant evals could elongate finiding the mate. The dll provides you just that so it CAN'T be any worse.

It can certainly be worse if the engine uses bitbases within the tree instead of in the leaves. You're not going to search the same tree and the effective depth reached is likely lower than what the engine would otherwise have reached. (And if you probe at the leaves instead of within the tree, you're going to lose most of the advantage of bitbases.)

It can also very easily be worse if there is any possibility of evaluation scores being compared to heuristic scores returned by your dll.

You are desperate and pathethic with all these non-existent if if. Yes they could fail if the earth were flat...

Your claim that the dll eval can be worse than the engine's evaluation is plain stupid and I have proved it to you. No body said anything about how the engine searches, infact already cautioned it can fail at low search depths.

My engine probes inside tree, after captures and promotions...
My engine compares eval with Bitbases scores, and thus prefers EGBB wins...
My enigine does everything with EGBBs.

So one more time please stop spreading LIES and go back to your Syzygy threads.

[syzygy]

You are desperate and pathethic with all these non-existent if if. Yes they could fail if the earth were flat...

You're as receptive for constructive criticism as ever.

[Daniel Shawul]

You are desperate and pathethic with all these non-existent if if. Yes they could fail if the earth were flat...

You're as receptive for constructive criticism as ever.

Yes when it comes from You and your cos about my egbbs, I am very careful.
Houdart where is he btw

[Daniel Mehrmann]

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.

I must agree here!

It would be better if your egbb interface would only send integer for win, draw or lose instead of eval terms.

I recommend egbb values should be filtered and replaced by own eval terms or search bounds.

Regards
Daniel
[/code]

[Daniel Shawul]

I guess your answer will be that your dll returns heuristic values that are supposed to help. My reply is that the engine is likely to be better at searching and evaluating the position than your dll. In any event, there is no way that the bitbase data ("everything wins") can help.

I must agree here!

It would be better if your egbb interface would only send integer for win, draw or lose instead of eval terms.

I recommend egbb values should be filtered and replaced by own eval terms or search bounds.

Regards
Daniel
[/code]

You have no clue for someone who agrees with an exclamation. First it does send an _integer_. Second if you mean bay a +1,0,-1, you can simply interpret >0 as +1, <0 as -1, and =0 as Draw. There is your beloved WDL score! So now you can add the KBNK table in your engine, and get the same thing, when you could have used what the dll provided you as an adjusted score... There is even more help coming to facilitate conversion, but if you want you not use it a ONE line code would do it. I for one am never going to change anything to fit your one line needs...

[hgm]

Indeed, that is how I use it in WinBoard. I am only interested in whether it is draw or not, so I just test whether the returned value equals 0.

[Max]

The use of bitbases likely makes it harder to win KBNK. If they are probed during the searched, it is overhead in return for no information. If they are used to prune branches, it is likely the search won't be able to make progress where without pruning it could. When KBNK is on the board, bitbases should be disabled.

Sorry Daniel, but Scorpio agreed here with Ronald.

Look at these two games with 7 sec/move on my core i7. In the first game (without egbbs) much deeper plys are searched and Scorpio won KBNK against a perfect playing opponent.

Code: Select all

[Event "Computer Chess Game"]
[Site "Home"]
[Date "2014.02.02"]
[Round "-"]
[White "Arasan 16.3 (Gaviota)"]
[Black "Scorpio 2014 (no Egbb)"]
[Result "0-1"]
[TimeControl "7sec/move"]
[FEN "8/7b/7n/4k3/2K5/8/8/8 w - - 0 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . b
. . . . . . . n
. . . . k . . .
. . K . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
white to play
--------------}
 1. Kc3 {-99,48/3} Nf5 {+13,67/23 7} 
 2. Kd3 {-99,48/3 0,3} Ng3+ {+13,70/25}
 3. Kc3 {-99,50/3} Be4 {+13,79/27} 
 4. Kb3 {-99,50/3 0,1} Kd4 {+13,79/29} 
 5. Kb2 {-99,52/3} Ne2 {+13,79/29} 
 6. Kb3 {-99,54/3} Kc5 {+13,79/28} 
 7. Ka2 {-99,54/3} Bd3 {+13,79/31} 
 8. Kb3 {-99,52/3} Kb5 {+13,79/26} 
 9. Ka2 {-99,54/3} Nd4 {+13,79/27} 
10. Ka1 {-99,56/3 0,1} Bg6 {+13,79/31} 
11. Kb2 {-99,56/3} Kc4 {+13,79/29} 
12. Ka2 {-99,58/3} Be4 {+13,79/29} 
13. Kb2 {-99,58/3 0,1} Nc6 {+172,00/28} 
14. Ka1 {-99,56/3} Kc3 {+173,60/31} 
15. Ka2 {-6,74/2} Bg6 {+174,40/31} 
16. Ka1 {-99,60/3 0,1} Bf7 {+99,38/29} 
17. Kb1 {-6,66/2} Nb4 {+99,40/29} 
18. Ka1 {-99,60/3} Kb3 {+99,42/29} 
19. Kb1 {-6,54/2} Nd3 {+99,44/28} 
20. Ka1 {-6,58/2 0,1} Bh5 {+99,46/29} 
21. Kb1 {-6,66/2} Bg6 {+99,48/29} 
22. Ka1 {-6,54/2} Ka3 {+99,50/29} 
23. Kb1 {-6,66/2} Ne1+ {+99,52/31} 
24. Ka1 {-99,58/3} Nc2+ {+99,54/33} 
25. Kb1 {-99,60/2} Nd4+ {+99,56/33} 
26. Kc1 {-99,62/3} Nf3 {+99,60/33} 
27. Kd1 {-6,98/2} Bd3 {+99,62/33} 
28. Kc1 {-6,98/2} Kb3 {+99,64/34} 
29. Kd1 {-7,06/2} Kc3 {+99,68/34} 
30. Kc1 {-6,98/2} Nd4 {+99,70/33} 
31. Kd1 {-7,02/2 0,1} Bc2+ {+99,72/33} 
32. Ke1 {-99,74/3} Kd3 {+99,74/32} 
33. Kf2 {-99,76/3} Bd1 {+99,76/32} 
34. Kg3 {-99,78/3} Ke3 {+99,78/30} 
35. Kh4 {-99,80/3} Kf4 {+99,80/31} 
36. Kh3 {-7,14/2} Bf3 {+99,82/32} 
37. Kh4 {-99,84/3 0,1} Nf5+ {+99,84/31} 
38. Kh3 {-99,86/2} Ke3 {+99,86/35} 
39. Kh2 {-7,22/2} Kf2 {+99,88/39} 
40. Kh3 {-7,10/2} Be2 {+99,90/55} 
41. Kh2 {-7,26/2} Bf1 {+99,92/64} 
42. Kh1 {-7,30/2} Nd4 {+99,94/64} 
43. Kh2 {-7,42/2} Nf3+ {+99,96/64} 
44. Kh1 {-99,98/2} Bg2# {+99,98/64}
{Xboard adjudication: Checkmate} 0-1

In the second game with egbbs, Scorpio managed only to draw.

Code: Select all

[Event "Computer Chess Game"]
[Site "Home"]
[Date "2014.02.02"]
[Round "-"]
[White "Arasan 16.3 (Gaviota)"]
[Black "Scorpio 2014"]
[Result "1/2-1/2"]
[TimeControl "7sec/move"]
[FEN "8/7b/7n/4k3/2K5/8/8/8 w - - 0 1"]
[SetUp "1"]

{--------------
. . . . . . . .
. . . . . . . b
. . . . . . . n
. . . . k . . .
. . K . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
white to play
--------------}
 1. Kc3 {-99,48/3} Ng4 {+46,04/21 7} 
 2. Kc4 {-99,50/3 0,3} Nf6 {+45,96/21}
 3. Kb3 {-99,50/3} Kd4 {+46,06/22} 
 4. Kb2 {-99,52/3 0,1} Ne4 {+45,92/22} 
 5. Kb3 {-99,54/3} Nc5+ {+45,89/22} 
 6. Kb2 {-99,56/3} Bf5 {+45,51/22} 
 7. Ka2 {-99,58/3} Kc3 {+45,74/22} 
 8. Ka1 {-99,60/3} Nb3+ {+45,76/22} 
 9. Ka2 {-99,62/2} Na5 {+45,39/23} 
10. Ka1 {-99,60/3} Be6 {+45,39/23} 
11. Kb1 {-6,30/2} Kd4 {+45,06/23} 
12. Kb2 {-99,54/3} Kc5 {+45,71/22} 
13. Kc2 {-99,52/3} Nc4 {+45,80/22} 
14. Kb3 {-99,52/3} Ne3+ {+45,45/22} 
15. Kc3 {-99,54/3} Ng2 {+45,45/22} 
16. Kd3 {-99,50/3} Kd6 {+45,45/23} 
17. Kd4 {-99,46/3} Bg4 {+45,73/22} 
18. Ke4 {-99,44/3 0,1} Ke6 {+45,33/22} 
19. Kd4 {-99,46/3} Nf4 {+45,71/22} 
20. Ke4 {-99,46/3} Nd5 {+45,35/22} 
21. Kd4 {-99,48/3} Be2 {+45,37/22} 
22. Ke4 {-99,46/3} Nb6 {+44,95/23} 
23. Kd4 {-99,44/3} Bf3 {+44,90/23} 
24. Ke3 {-99,46/3} Ba8 {+44,52/24} 
25. Kd4 {-99,44/3 0,1} Nd7 {+44,95/23} 
26. Kc3 {-99,46/3} Kd5 {+45,14/23} 
27. Kd3 {-99,48/3} Nf6 {+44,97/23} 
28. Kc3 {-99,48/3} Ke4 {+44,97/23} 
29. Kb2 {-99,50/3} Kf3 {+44,80/23} 
30. Ka1 {-99,48/3} Ne4 {+44,97/21} 
31. Kb1 {-99,50/3 0,1} Nf2 {+44,07/21} 
32. Ka1 {-99,48/3} Ke3 {+44,84/23} 
33. Kb2 {-99,50/3} Kd3 {+45,06/23} 
34. Kb3 {-99,52/3} Bd5+ {+45,05/23} 
35. Kb2 {-99,54/3 0,1} Nd1+ {+45,49/23} 
36. Ka1 {-99,56/3} Nc3 {+45,31/23} 
37. Kb2 {-6,90/2} Bg8 {+45,27/22} 
38. Kc1 {-99,54/3 0,1} Na4 {+44,89/24} 
39. Kb1 {-99,56/3} Kd2 {+43,81/25} 
40. Ka1 {-6,22/2} Kc3 {+44,46/23} 
41. Kb1 {-6,30/2} Nb2 {+45,26/21} 
42. Ka1 {-99,56/3} Nd3 {+44,11/25} 
43. Kb1 {-6,66/2} Nb4 {+43,62/25} 
44. Ka1 {-99,60/3} Nd5 {+43,52/26} 
45. Kb1 {-99,58/3} Ne3 {+0,00/47} 
46. Ka1 {-99,60/3} Nf1 {+0,00/64} 
47. Kb1 {-6,50/2} Bb3 {+0,00/64} 
48. Ka1 {-99,60/3 0,1} Ng3 {-0,02/64} 
49. Kb1 {-6,58/2} Bg8 {-0,02/64} 
50. Ka1 {-99,56/3 0,1} Kd4 {-0,02/64}
{Draw claim: 50-move rule} 1/2-1/2

-Max

[Daniel Shawul]

Scorpio can't win anything without egbbs because it _doesn't_ have the KBNK table. How the hell it manage to win without it? I already posted a game that showed that it can't win it. Even with EGBBs it may not be able to because of prunings that I may have forgotten to disable. I haven't even released a Scorpio version yet, so why do you bother me with something I have not released yet ?? Play Stockfish EGBB against it and see if it is a depth/pruning problem because I am sure stockfish disables most of it with abs(score) < WIN_SCORE stuff. So stop lecturing me on what has worked for years.