With Acid-Ape go to Engine Settings, Scroll down and make sure you have the following filled in:
Working Directory
/Use/Full/Path to directory
WeightsFile
/Use/Full/Path to file
Syzygypath
/Use/Full/Path to directory
Moderators: hgm, Rebel, chrisw
Thanks, it worksabik wrote: ↑Fri May 24, 2019 7:18 pmActually you can run LC0 in Chess for Android too with an extra setup step.After this you can play lc0. The auto-discover default will find the weights file in the same directory.
- Copy lc0 binary and weights files to /sdcard
- Install engine from SD card on lc0
- Install engine from SD card on weights file (not intuitive, but this puts file in right place)
- Import engine lc0
CDrill 1800 with around 1800 rating on CCRL 40/4 can beat Lc0 16x2, I think it is normal as Bikjump is way stronger than Cdrill. Net 48x5 is better than 16x2 based from my test on my tab. Also 48x5 is better than 64x6. But 32x4 is doing good vs 48x5, these of course varies from devices to devices. These tests are based only on a 4 or more game test matches using Acid Ape Chess at TC 3m+2s.abik wrote: ↑Fri May 24, 2019 8:10 pm I did a very quick sanity 1-second-per-move tournament between BikJump and LC0 (weights file 11258-16x2-se-4.pb) in Chess for Android, but that did not go well for LC0. Since I (unfortunately) know that BikJump is not very strong, I must be doing something wrong still. Is there something essential in the setup that I am missing?
Code: Select all
Head to head statistics:
1) Lc0 v0.22.0-dev w11258-48x5 1907.9 : 27 (+10,=12,-5), 59.3 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
Lc0 v0.22.0-dev w11258-32x4 : 8 ( 2, 4, 2), 50.0 : +0.0, 106.3, 50.0
Lc0 v0.22.0-dev w11258-64x6 : 15 ( 6, 7, 2), 63.3 : +103.8, 77.2, 91.1
CDrill 1800 : 4 ( 2, 1, 1), 62.5 : +107.9, 147.0, 76.9
2) Lc0 v0.22.0-dev w11258-32x4 1907.9 : 8 (+2,=4,-2), 50.0 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
Lc0 v0.22.0-dev w11258-48x5 : 8 ( 2, 4, 2), 50.0 : -0.0, 106.3, 50.0
3) Lc0 v0.22.0-dev w11258-64x6 1804.2 : 15 (+2,=7,-6), 36.7 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
Lc0 v0.22.0-dev w11258-48x5 : 15 ( 2, 7, 6), 36.7 : -103.8, 77.2, 8.9
4) CDrill 1800 1800.0 : 4 (+1,=1,-2), 37.5 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
Lc0 v0.22.0-dev w11258-48x5 : 4 ( 1, 1, 2), 37.5 : -107.9, 147.0, 23.1
Thanks. I am currently using a Pixel C.Ferdy wrote: ↑Fri May 24, 2019 8:41 pmCDrill 1800 with around 1800 rating on CCRL 40/4 can beat Lc0 16x2, I think it is normal as Bikjump is way stronger than Cdrill. Net 48x5 is better than 16x2 based from my test on my tab. Also 48x5 is better than 64x6. But 32x4 is doing good vs 48x5, these of course varies from devices to devices. These tests are based only on an 8-game test matches using Acid Ape Chess.
I only have BikJump 2.1 for Android from your website. On one core of my phone it gets about 250-300knps. Using Chess for Android, Lc0 11258-48x5-se.pb on 4 cores beat it +7 -1 =2 in games at 10s/move. I would estimate BikJump on 1 core of my phone as about 2000 CCRL 40/4, or maybe 2200 FIDE rating in Blitz (maybe even 2300). Therefore Lc0 on 4 cores is about 2400-2500 FIDE level in blitz using this net.abik wrote: ↑Fri May 24, 2019 8:46 pmThanks. I am currently using a Pixel C.Ferdy wrote: ↑Fri May 24, 2019 8:41 pmCDrill 1800 with around 1800 rating on CCRL 40/4 can beat Lc0 16x2, I think it is normal as Bikjump is way stronger than Cdrill. Net 48x5 is better than 16x2 based from my test on my tab. Also 48x5 is better than 64x6. But 32x4 is doing good vs 48x5, these of course varies from devices to devices. These tests are based only on an 8-game test matches using Acid Ape Chess.
A quick 2-seconds-per-move tournament using the weights file 11258-48x5-se.pb tipped the balance, 5.5 vs 4.5 in LC0's favor.
Here is a sample game of one of LC0's victories (per adjudication).
[pgn][Event "Chess for Android Tournament"] [Site "United States"] [Date "2019.05.24"] [Round "10"] [White "Lc0 v0.22.0-dev"] [WhiteElo "-"] [Black "BikJump v2.5 (32-bit)"] [BlackElo "-"] [Result "1-0"] [PlyCount "109"] [TimeControl "1/2"] [Device "Pixel C"] 1. d4 {opening reply 0.001} Nf6 {opening reply 0.0} 2. c4 {opening reply 0.0} e6 {opening reply 0.0} 3. g3 {opening reply 0.0} d5 {opening reply 0.0} 4. Bg2 {opening reply 0.0} dxc4 {opening reply 0.0} 5. Qa4+ {[5/10]8/b8d7 1.684} c6 {[9/9]-3/a4c4 2.072} 6. Qxc4 {[6/11]14/b7b5 1.155} Be7 {[8/16]-8/b1c3 2.05} 7. Nf3 {[5/11]15/e8g8 1.263} Ng4 {[8/16]-4/c1f4 2.039} 8. O-O {[5/9]28/e8g8 1.666} b5 {[7/14]-20/c4c2 1.452} 9. Qc2 {[5/10]41/c8b7 1.264} Bb7 {[8/16]-29/h2h3 2.0} 10. a4 {[5/10]38/e8g8 1.76} b4 {[7/14]-32/h2h3 1.773} 11. a5 {[5/9]48/e8g8 1.656} Qd5 {[8/8]-29/h2h3 2.009} 12. Bf4 {[4/9]85/e8g8 1.765} b3 {[7/14]-57/c2c3 2.089} 13. Qd1 {[5/8]142/c6c5 1.788} Qc4 {[7/14]-62/h2h3 1.725} 14. Nbd2 {[5/9]144/c4c2 1.605} Qc2 {[9/18]-92/h2h3 2.101} 15. Nxb3 {[6/11]183/c2d1 1.315} Qxd1 {[8/16]-131/f1d1 2.067} 16. Rfxd1 {[6/13]170/b7a6 1.314} O-O {[8/16]-137/f3e5 1.639} 17. Nc5 {[6/12]211/b7a6 1.504} Bc8 {[9/18]-137/h2h3 1.434} 18. Rac1 {[4/11]240/g4f6 1.81} Kh8 {[9/18]-138/h2h3 2.015} 19. h3 {[4/9]332/g4f6 1.811} Nf6 {[10/10]-143/h3h4 2.024} 20. Ne5 {[5/10]322/f6d5 1.306} Nd5 {[9/18]-139/c5d3 1.4} 21. Bd2 {[5/10]298/f7f6 1.264} f6 {[8/16]-143/e5f3 1.468} 22. Ned3 {[5/11]259/h8g8 1.163} g5 {[8/16]-139/b2b4 2.105} 23. e4 {[4/9]269/d5c7 1.77} Nc7 {[9/18]-136/d3b4 1.793} 24. e5 {[4/9]310/f6f5 1.571} Nd5 {[7/14]-139/g2e4 1.213} 25. exf6 {[5/8]419/e7f6 1.553} Bxf6 {[10/20]-157/d3e5 2.016} 26. Ne5 {[5/9]438/b8a6 1.571} Bxe5 {[10/20]-176/d4e5 2.025} 27. dxe5 {[5/10]431/h7h6 1.134} h6 {[9/18]-188/g2e4 1.315} 28. Ne4 {[5/9]314/b8d7 1.811} Nd7 {[10/20]-140/h3h4 2.049} 29. Rxc6 {[5/10]211/d7e5 1.673} Nxe5 {[10/10]-126/c6c5 2.044} 30. Bc3 {[6/11]210/d5c3 1.816} Nxc3 {[10/20]-128/c6c3 1.452} 31. Rxc3 {[6/11]214/c8b7 0.973} Rb8 {[9/18]-140/c3c7 1.792} 32. Rc5 {[6/10]267/e5g6 1.264} Nf7 {[10/20]-144/d1d2 1.998} 33. Rd2 {[5/10]303/c8b7 1.76} e5 {[10/20]-166/e4f6 2.133} 34. Rc7 {[5/9]354/a7a6 1.265} a6 {[11/22]-248/e4d6 6.608} 35. Nf6 {[5/11]333/h8g7 1.81} Bf5 {[9/18]-192/c7a7 1.417} 36. g4 {[5/10]335/f5e6 1.759} Rbc8 {[11/22]-238/c7e7 3.832} 37. Ra7 {[5/10]348/f5g6 1.803} Rc1+ {[10/20]-255/g1h2 2.05} 38. Kh2 {[5/10]363/f5g6 1.247} Be6 {[10/20]-259/a7a6 2.036} 39. Rxa6 {[5/9]522/f7d8 1.345} Bc8 {[10/10]-267/a6c6 2.013} 40. Ra8 {[5/9]559/h8g7 1.761} Rc7 {[9/23]-306/a5a6 2.016} 41. a6 {[4/9]696/h8g7 1.809} Kg7 {[10/20]-439/f6h5 2.033} 42. Nh5+ {[5/11]953/g7h7 1.365} Kh8 {[10/20]-467/a6a7 1.56} 43. a7 {[4/10]932/e5e4 1.76} Nd8 {[10/21]-499/a8b8 2.062} 44. Rb8 {[4/10]945/c7a7 1.807} Rxa7 {[11/22]-648/b8c8 1.35} 45. Rxc8 {[5/8]856/d8e6 1.161} Kg8 {[10/25]-955/c8d8 1.862} 46. Rdxd8 {[4/9]2489/f8d8 1.776} Rxd8 {[12/12]-961/c8d8 2.006} 47. Rxd8+ {[4/7]1263/g8f7 1.187} Kf7 {[12/24]-995/d8d6 1.217} 48. Ng3 {[4/7]890/a7a1 1.809} Ke7 {[10/21]-974/d8d5 2.003} 49. Rd1 {[4/7]859/a7a2 1.818} Ra6 {[10/23]-970/g3f5 1.486} 50. Nf5+ {[3/7]880/e7f7 1.808} Kf6 {[11/25]-989/f5h6 2.002} 51. Rd8 {[4/8]999/a6a7 1.809} Rb6 {[10/27]-1030/d8d6 2.029} 52. Bd5 {[3/7]1361/b6a6 1.812} e4 {[10/26]-1054/d8e8 1.468} 53. Bxe4 {[3/6]1003/f6e5 1.827} Rxb2 {[10/23]-998/h2g1 1.418} 54. Kg3 {[3/7]962/b2b3 1.775} Rb6 {[10/22]-1009/f5h6 2.025} 55. Bd5 1-0 {black resigned:adjudication} [/pgn]
Thank you for those data points. I am now running a 5 minutes for 40 moves gauntlet between LC0 and some other Android engines. I will post the results when available.Laskos wrote: ↑Fri May 24, 2019 10:44 pmI only have BikJump 2.1 for Android from your website. On one core of my phone it gets about 250-300knps. Using Chess for Android, Lc0 11258-48x5-se.pb on 4 cores beat it +7 -1 =2 in games at 10s/move. I would estimate BikJump on 1 core of my phone as about 2000 CCRL 40/4, or maybe 2200 FIDE rating in Blitz (maybe even 2300). Therefore Lc0 on 4 cores is about 2400-2500 FIDE level in blitz using this net.
It also managed to get 3 draws in 10 games against Ethereal 9.00 one core, which is about 2800 CCRL Elo level engine on my phone, in line with 2400-2500 FIDE rating of Lc0 11258-48x5-se.pb on my phone.
DroidFish has an old optimization that makes it not send the complete game history to the chess engine. Moves in the history that cannot affect the game theoretical value of the current game are not sent. This is bad in the lc0 case since it will cause the history planes to not get set up correctly.
Interesting game! LC0 working in Acid Ape chess Arm8.....Ferdy wrote: ↑Fri May 24, 2019 7:25 am Tried the 64bit compile and it worked. I use Acid Ape Chess Grandmaster Edition to run the engine at 1 core, on blas backend on Android 7.0. Here is a sample game using network file 11258-16x2-se from dkappe. Speed is around 500 nps from startpos.
Weak at tactics but plays an opening as if it is using an opening book.
[pgn] [Event "3+2 engine game"] [Site "?"] [Date "2019.05.24"] [Round "?"] [White "Myself"] [Black "Lc0 v0.22.0-dev"] [Result "1-0"] [WhiteCountry "usa"] [BlackElo "3049"] [TimeControl "3+2"] [Time "12:40:28"] [AacUser "White"] [AacWhitePlayerType "Human"] [AacBlackPlayerType "Engine"] [AacIsRated "0"] [AacDuration "04:07.232"] 1. d4 d5 {[%clk 02:58]} 2. c4 {[%clk 02:59]} c6 {[%clk 02:56]} 3. e3 {[%clk 02:40]} Nf6 {[%clk 02:55]} 4. Nc3 {[%clk 02:38]} e6 {[%clk 02:53]} 5. Qc2 {[%clk 02:34]} Nbd7 {[%clk 02:51]} 6. Nf3 {[%clk 02:33]} Bd6 {[%clk 02:50]} 7. Be2 {[%clk 02:27]} O-O {[%clk 02:48]} 8. O-O {[%clk 02:27]} dxc4 {[%clk 02:46]} 9. Bxc4 {[%clk 02:26]} b5 {[%clk 02:44]} 10. Bd3 {[%clk 02:24]} Bb7 {[%clk 02:42]} 11. Rd1 {[%clk 02:22]} a6 {[%clk 02:40]} 12. e4 {[%clk 02:12]} c5 {[%clk 02:39]} 13. e5 {[%clk 02:02]} Bxf3 {[%clk 02:38]} 14. gxf3 {[%clk 02:00]} cxd4 {[%clk 02:37]} 15. exd6 {[%clk 01:50]} dxc3 {[%clk 02:35]} 16. Qxc3 {[%clk 01:50]} Rc8 {[%clk 02:34]} 17. Qd4 {[%clk 01:48]} Nd5 {[%clk 02:31]} 18. Bd2 {[%clk 01:43]} Rc6 {[%clk 02:29]} 19. Qe4 {[%clk 01:37]} N7f6 {[%clk 02:26]} 20. Qh4 {[%clk 01:33]} Rxd6 {[%clk 02:23]} 21. Bg5 {[%clk 01:30]} h6 {[%clk 02:21]} 22. Bxh6 {[%clk 01:28]} gxh6 {[%clk 02:19]} 23. Qxh6 {[%clk 01:27]} Re8 {[%clk 02:17]} 24. Kh1 {[%clk 01:27]} Nb4 {[%clk 02:14]} 25. Rg1+ {[%clk 01:24]} Ng4 {[%clk 02:16]} 26. Rxg4+ {[%clk 01:23]} Qg5 {[%clk 02:18]} 27. Rxg5+ {[%clk 01:21]} 1-0 [/pgn]
Tried to intall it at different gui.
1. Droidfish, ok
2. Chess for android, not ok
3. Analyze this pro, not ok
4. Acid Ape Chess Grandmaster edition, ok
Items 2 and 3 did not work because the WeightsFile option cannot be set.
Auto discover of weight file does not work, it should be set via WeightsFile option which items 1 and 4 are capable.