29 Alpha-Beta's against Leela...

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
zullil
Posts: 5601
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: 29 Alpha-Beta's against Leela...

Post by zullil » Wed Aug 28, 2019 9:02 am

peter wrote:
Wed Aug 28, 2019 5:01 am

And here I did let SF run over night at 24...Rec8 again.

Analysis by Brainfish 210819 64 POPCNT:


Code: Select all

25.Sa3 c3 26.Sc2 Sb2 27.a3 h5 28.f3 hxg4 29.hxg4 Sc4 30.Kh2 f6 31.Dd1 g5 32.Lg3 Sxa3 33.f4 Le8 34.fxg5 Sxc2 35.Dxc2 Ta2 36.Dxa2 c2 37.Dxd5+ Lf7 38.Dxb5 c1D 39.Df5 Tc2+ 40.Lf2 Tb2 41.Kg3 Dc7+ 42.Kf3 Dc6+ 43.Kg3 fxg5 44.Dxg5+ Lg6 45.d5 De8 46.d6 Td2 47.Le1 Txd6 48.Lc3 Te6 49.Dd5 De7 50.Le5 Le4 51.Dxe4 Txe5 52.Dg6+ Kf8 53.Dh6+ Ke8 54.Kf4 Kd7 55.Db6 Te4+ 56.Kf3 Txb4 57.Da7+ Ke6 58.Dxe7+ Kxe7 59.g5 Tb1 60.Kg2 Te1 
  -+  (-3.19)   Tiefe: 53/91   06:17:36  470463MN
25.Sa3 c3 
  -+  (-3.30 --)   Tiefe: 54/66   06:38:43  496967MN
24 threads of 12x3GHz, 32G hash, no tbs, started with already full hash after forwarding and backwarding the known best lines.

After 2 hours of switching, 25.Nd2 doesn't come up anymore.
Na3 is kept best for the last four and a half hours except a single line- try with Nc3 after 5 hours.

So 25.Na3 might be best defence for both first moves yet, for 24...c3 as well as for 24...Rec8, latter one just quite as well as the first one therefore as 24th move of Black's, both transposing to each other 2 plies later on.
I also wondered if 25.Na3 might be better than 25.Nd2, so I also ran an overnight search with Stockfish-dev, using an already full hash and MultiPV = 2:

-3.39 2. Nd2 h5 3. gxh5 c3 4. Nb3 c2 5. Kh2 Nc3 6. Qg1 Ne2 7. Qg5 Nxf4 8. exf4 Ra6 9. Nc1 Rf6 10. Qxd5 Be6 11. Qf3 Rc4 12. Qe3 Rf5 13. h6 gxh6 14. Qg3+ Kh8 15. Qe3 Rh5 16. Qe2 Rd5 17. a4 Rdxd4 18. a5 Rd1 19. Qe3 Kg8 20. f3 Bd5 21. a6 Rxb4 22. a7 Ra4 23. f5 Kh7 24. f6 Kg6 25. Ne2 Rc4 26. Kg3 c1=Q 27. Nxc1 Rcxc1 28. Qa3 Ba8 29. Kh4 Rd4+ 30. Kg3 Rg1+ 31. Kh2 Rdd1 32. Qf8 Bxf3 33. a8=Q (depth 53, 9:59:22)

-3.57 2. Na3 c3 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Bg3 f6 8. Be1 Ra6 9. a4 bxa4 10. b5 Ra5 11. Bxc3 Rxb5 12. Nb4 a3 13. Qa1 Ra8 14. Qa2 Nxe3 15. Qe2 a2 16. Nxa2 Rb1+ 17. Kf2 Rb3 18. Qd2 Nd1+ 19. Kg1 Rb1 20. Kh2 Ra3 21. Qc2 Rbb3 22. Bb4 Rxa2 23. Qxa2 Rb2+ 24. Qxb2 Nxb2 25. Bd2 g5 26. Kg2 Nd3 27. Kf1 Ba4 28. Ke2 Nf4+ 29. Kf2 Bd1 30. Ba5 Ng6 31. Bd2 Nh4 32. f4 Nf3 33. Be3 gxf4 34. Bxf4 Nxd4 35. Bc1 Ne6 36. Bb2 Kf7 (depth 53, 9:59:22)

At the moment, Stockfish prefers Nd2 again. In any case, it seems that 24...Rec8 is nearly as strong as 24...c3.

Fulvio
Posts: 145
Joined: Fri Aug 12, 2016 6:43 pm

Re: 29 Alpha-Beta's against Leela...

Post by Fulvio » Wed Aug 28, 2019 9:13 am

Glarean wrote:
Mon Aug 26, 2019 8:26 am
I organized a engines tournament; the A/B's got their best conditions, and the NN's got their best conditions. At the end I added up - basta :-)
Great article, very interesting.
To give every engine the best condition, I think it would be preferable to do two matches for each position, with inverted colors.
For example the position: r1bqkbnr/pp3ppp/2npp3/8/2PNP3/8/PP3PPP/RNBQKB1R w KQkq - 0 6
SharedScreenshot.jpg
SharedScreenshot.jpg (181.68 KiB) Viewed 845 times
is very interesting: in the July Lichess' database after Nc3 white won 58% of the games and lost 36% of them.
However in your tournament Stockfish played twice as black :(

zullil
Posts: 5601
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: 29 Alpha-Beta's against Leela...

Post by zullil » Wed Aug 28, 2019 8:39 pm

zullil wrote:
Wed Aug 28, 2019 9:02 am
peter wrote:
Wed Aug 28, 2019 5:01 am

And here I did let SF run over night at 24...Rec8 again.

Analysis by Brainfish 210819 64 POPCNT:


Code: Select all

25.Sa3 c3 26.Sc2 Sb2 27.a3 h5 28.f3 hxg4 29.hxg4 Sc4 30.Kh2 f6 31.Dd1 g5 32.Lg3 Sxa3 33.f4 Le8 34.fxg5 Sxc2 35.Dxc2 Ta2 36.Dxa2 c2 37.Dxd5+ Lf7 38.Dxb5 c1D 39.Df5 Tc2+ 40.Lf2 Tb2 41.Kg3 Dc7+ 42.Kf3 Dc6+ 43.Kg3 fxg5 44.Dxg5+ Lg6 45.d5 De8 46.d6 Td2 47.Le1 Txd6 48.Lc3 Te6 49.Dd5 De7 50.Le5 Le4 51.Dxe4 Txe5 52.Dg6+ Kf8 53.Dh6+ Ke8 54.Kf4 Kd7 55.Db6 Te4+ 56.Kf3 Txb4 57.Da7+ Ke6 58.Dxe7+ Kxe7 59.g5 Tb1 60.Kg2 Te1 
  -+  (-3.19)   Tiefe: 53/91   06:17:36  470463MN
25.Sa3 c3 
  -+  (-3.30 --)   Tiefe: 54/66   06:38:43  496967MN
24 threads of 12x3GHz, 32G hash, no tbs, started with already full hash after forwarding and backwarding the known best lines.

After 2 hours of switching, 25.Nd2 doesn't come up anymore.
Na3 is kept best for the last four and a half hours except a single line- try with Nc3 after 5 hours.

So 25.Na3 might be best defence for both first moves yet, for 24...c3 as well as for 24...Rec8, latter one just quite as well as the first one therefore as 24th move of Black's, both transposing to each other 2 plies later on.
I also wondered if 25.Na3 might be better than 25.Nd2, so I also ran an overnight search with Stockfish-dev, using an already full hash and MultiPV = 2:

-3.39 2. Nd2 h5 3. gxh5 c3 4. Nb3 c2 5. Kh2 Nc3 6. Qg1 Ne2 7. Qg5 Nxf4 8. exf4 Ra6 9. Nc1 Rf6 10. Qxd5 Be6 11. Qf3 Rc4 12. Qe3 Rf5 13. h6 gxh6 14. Qg3+ Kh8 15. Qe3 Rh5 16. Qe2 Rd5 17. a4 Rdxd4 18. a5 Rd1 19. Qe3 Kg8 20. f3 Bd5 21. a6 Rxb4 22. a7 Ra4 23. f5 Kh7 24. f6 Kg6 25. Ne2 Rc4 26. Kg3 c1=Q 27. Nxc1 Rcxc1 28. Qa3 Ba8 29. Kh4 Rd4+ 30. Kg3 Rg1+ 31. Kh2 Rdd1 32. Qf8 Bxf3 33. a8=Q (depth 53, 9:59:22)

-3.57 2. Na3 c3 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Bg3 f6 8. Be1 Ra6 9. a4 bxa4 10. b5 Ra5 11. Bxc3 Rxb5 12. Nb4 a3 13. Qa1 Ra8 14. Qa2 Nxe3 15. Qe2 a2 16. Nxa2 Rb1+ 17. Kf2 Rb3 18. Qd2 Nd1+ 19. Kg1 Rb1 20. Kh2 Ra3 21. Qc2 Rbb3 22. Bb4 Rxa2 23. Qxa2 Rb2+ 24. Qxb2 Nxb2 25. Bd2 g5 26. Kg2 Nd3 27. Kf1 Ba4 28. Ke2 Nf4+ 29. Kf2 Bd1 30. Ba5 Ng6 31. Bd2 Nh4 32. f4 Nf3 33. Be3 gxf4 34. Bxf4 Nxd4 35. Bc1 Ne6 36. Bb2 Kf7 (depth 53, 9:59:22)

At the moment, Stockfish prefers Nd2 again. In any case, it seems that 24...Rec8 is nearly as strong as 24...c3.
Stopping here. Stockfish still sees Nd2 as the better of the two, though both are losing ...

-3.52 2. Nd2 h5 3. gxh5 c3 4. Nb3 c2 5. Kh2 Nc3 6. Be5 Rc6 7. Qg1 f6 8. Bf4 Bf5 9. h6 g6 10. e4 Nxe4 11. Qf1 Rc4 12. Bc1 Ra7 13. h4 Kh7 14. Qe2 Nc3 15. Qe8 Rc8 16. Qe3 Nxa2 17. Bd2 Rc4 18. Qf4 Rf7 19. h5 Nc3 20. Nc1 Ne4 21. Be3 Rxb4 22. f3 Nc3 23. Qb8 Rc4 24. hxg6+ Bxg6 25. Qb6 Rb4 26. Qa5 Rb1 27. Qxc3 b4 28. Qc6 b3 29. Nxb3 Rxb3 30. Qe6 Rc3 31. Kg3 c1=Q 32. Bxc1 Rxc1 33. Qxd5 Rc3 34. Kf2 Kxh6 35. f4 Rh3 36. f5 Bh5 37. Qe6 Kg7 38. Qe4 Bf3 (depth 56, 22:08:42)

-3.73 2. Na3 c3 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Qa1 Be6 8. Qc1 f6 9. e4 dxe4 10. fxe4 Bxg4 11. Qb1 Kf7 12. Qf1 Rh8 13. e5 Rad8 14. Qg2 Rh4 15. Bg3 Rh5 16. Qb7+ Rd7 17. Qxb5 Be6 18. Be1 fxe5 19. d5 Rg5+ 20. Kf2 Rxd5 21. Qxc4 Rd2+ 22. Bxd2 Rf5+ 23. Kg3 Bxc4 24. Bxc3 Bd5 25. Ne1 Bc6 26. Bd2 Ke6 27. Be3 Kd5 28. Bc5 Kc4 29. a4 Bxa4 30. Ng2 Kd3 31. Ne3 Rf4 32. Ng4 Kc3 33. Ne3 (depth 56, 22:08:42)

Post Reply