| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Sun Sep 11 14:05:39 2022 +0200 Timestamp: 1662897939 Further LTC tuning of search parameters Tuning done by bigpenor with some hand adjustments on top by Viz. Had a good performance at fixed games 180+1.8: https://tests.stockfishchess.org/tests/view/631836b437f41b13973d7da1 Elo: 1.35 +-1.2 (95%) LOS: 98.6% Total: 60000 W: 16422 L: 16189 D: 27389 Elo +1.35 Ptnml(0-2): 39, 5335, 18992, 5622, 12 nElo: 3.13 +-2.8 (95%) PairsRatio: 1.05 Passed 60+0.6 8 threads SPRT: https://tests.stockfishchess.org/tests/view/631ba0ff74bc4fe483a99db3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 29712 W: 8301 L: 8039 D: 13372 Elo +3.06 Ptnml(0-2): 12, 2318, 9925, 2598, 3 closes https://github.com/official-stockfish/Stockfish/pull/4160 bench 3938073 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: FauziAkram
Date: Wed Sep 7 07:38:04 2022 +0200 Timestamp: 1662529084 VLTC tuning Tuning some parameters that scale well with longer time control: Failed STC: https://tests.stockfishchess.org/tests/view/6313424d8202a039920e130a LLR: -2.94 (-2.94,2.94) <-1.75,0.25> Total: 42680 W: 11231 L: 11540 D: 19909 Elo -2.52 Ptnml(0-2): 191, 5008, 11232, 4737, 172 Passed LTC: https://tests.stockfishchess.org/tests/view/6311e2cd874169ca52ae7933 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 53448 W: 14782 L: 14437 D: 24229 Elo +2.24 Ptnml(0-2): 101, 5214, 15740, 5577, 92 Passed VLTC: https://tests.stockfishchess.org/tests/view/6312530cfa99a92e3002c927 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 123336 W: 33465 L: 33007 D: 56864 Elo +1.29 Ptnml(0-2): 38, 11466, 38204, 11920, 40 closes https://github.com/official-stockfish/Stockfish/pull/4154 Bench: 5609606 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Wed Sep 7 07:32:53 2022 +0200 Timestamp: 1662528773 Provide network download fallback in case the base infrastructure for providing the networks https://tests.stockfishchess.org/nns is down, use an alternate github repo for downloading networks during the build. fixes #4149 fixes #4140 closes https://github.com/official-stockfish/Stockfish/pull/4151 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Wed Sep 7 07:31:48 2022 +0200 Timestamp: 1662528708 Increase the maximum number of threads to 1024 relatively soon servers with 512 threads will be available 'quite commonly', anticipate even more threads, and increase our current maximum from 512 to 1024. closes https://github.com/official-stockfish/Stockfish/pull/4152 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: dav1312
Date: Mon Aug 29 19:15:14 2022 +0200 Timestamp: 1661793314 Disable ARM CI tests Temporarily disable ARM CI tests until a mitigation is implemented closes https://github.com/official-stockfish/Stockfish/pull/4148 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: mstembera
Date: Wed Aug 17 19:56:15 2022 +0200 Timestamp: 1660758975 Make key_after() more consistent with key() STC: https://tests.stockfishchess.org/tests/view/62f8547123d42b50a8dac674 LLR: 2.95 (-2.94,2.94) <0.00,2.00> Total: 176640 W: 47699 L: 47189 D: 81752 Elo +1.00 Ptnml(0-2): 776, 18599, 49129, 18971, 845 A bug fix plus non functional speed optimization. Position::key_after(Move m) is now consistent with Position::key() thus prefetching correct TT entries which speeds things up. Related PR #3759 closes https://github.com/official-stockfish/Stockfish/pull/4130 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Mon Aug 15 18:01:37 2022 +0200 Timestamp: 1660579297 Simplify the use of classical eval no benefit of the fallback term (exercised rarely). Cleanup the associated code. passed STC https://tests.stockfishchess.org/tests/view/62f62c2b6f0a08af9f776367 LLR: 2.96 (-2.94,2.94) <-1.75,0.25> Total: 67832 W: 18334 L: 18148 D: 31350 Elo +0.95 Ptnml(0-2): 369, 7171, 18609, 7439, 328 passed LTC https://tests.stockfishchess.org/tests/view/62f68beb6f0a08af9f77710e LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 104664 W: 28363 L: 28233 D: 48068 Elo +0.43 Ptnml(0-2): 169, 10162, 31511, 10350, 140 closes https://github.com/official-stockfish/Stockfish/pull/4132 Bench: 6079565 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Mon Aug 15 17:54:56 2022 +0200 Timestamp: 1660578896 Simplify away smp adjustment in TT use Passed STC https://tests.stockfishchess.org/tests/view/62f7d81f23d42b50a8dab568 LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 98160 W: 26307 L: 26165 D: 45688 Elo +0.50 Ptnml(0-2): 201, 10282, 27960, 10448, 189 Passed LTC https://tests.stockfishchess.org/tests/view/62f8d1a623d42b50a8dad4fb LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 81544 W: 22346 L: 22200 D: 36998 Elo +0.62 Ptnml(0-2): 44, 7542, 25446, 7704, 36 closes https://github.com/official-stockfish/Stockfish/pull/4131 No functional change (single threaded). see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: mckx00
Date: Mon Aug 15 17:51:51 2022 +0200 Timestamp: 1660578711 Make LMR code easier to follow Remove flags doFullDepthSearch and didLMR, and reorder instruction. Small measured speedup. Closes https://github.com/official-stockfish/Stockfish/pull/4129 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: mstembera
Date: Mon Aug 15 17:46:27 2022 +0200 Timestamp: 1660578387 Report longest PV lines for multithreaded search In case several threads find the same bestmove, report the longest PV line found. closes https://github.com/official-stockfish/Stockfish/pull/4126 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Fri Aug 12 14:30:33 2022 +0200 Timestamp: 1660307433 Remove an unneeded randomization of evals. most of the effect comes from the randomization of 3-folds. passed STC: https://tests.stockfishchess.org/tests/view/62e697e97e84186e5d19af6f LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 572976 W: 153168 L: 153539 D: 266269 Elo -0.22 Ptnml(0-2): 2505, 64783, 152364, 64250, 2586 passed LTC: https://tests.stockfishchess.org/tests/view/62ee5977523c86dcd6957154 LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 704808 W: 191212 L: 191680 D: 321916 Elo -0.23 Ptnml(0-2): 1340, 70579, 208972, 70235, 1278 closes https://github.com/official-stockfish/Stockfish/pull/4128 Bench: 5868987 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Stefan Geschwentner
Date: Fri Aug 12 14:29:40 2022 +0200 Timestamp: 1660307380 Cleanup code This PR includes following cleanups: - Remove the unused depth variable in the thread class. - cleanup ValueList (added from mstembera) closes https://github.com/official-stockfish/Stockfish/pull/4127 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Sat Aug 6 13:57:30 2022 +0200 Timestamp: 1659787050 Update WDL model for current SF This updates the WDL model based on the LTC statistics for the two weeks (3M games). for old results see: https://github.com/official-stockfish/Stockfish/pull/3981 https://github.com/official-stockfish/Stockfish/pull/3582 https://github.com/official-stockfish/Stockfish/pull/2778 closes https://github.com/official-stockfish/Stockfish/pull/4115 No functional change. see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Sat Aug 6 13:53:55 2022 +0200 Timestamp: 1659786835 Update CPU contributors list Thanks for your contributions! closes https://github.com/official-stockfish/Stockfish/pull/4116 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: lonfom169
Date: Sat Aug 6 13:52:36 2022 +0200 Timestamp: 1659786756 Reintroduce singularQuietLMR STC: LLR: 2.96 (-2.94,2.94) <0.00,2.00> Total: 88912 W: 23972 L: 23580 D: 41360 Elo +1.53 Ptnml(0-2): 365, 9820, 23712, 10176, 383 https://tests.stockfishchess.org/tests/view/62e9537a400addce2c13399b LTC: LLR: 2.97 (-2.94,2.94) <0.50,2.50> Total: 85672 W: 23607 L: 23192 D: 38873 Elo +1.68 Ptnml(0-2): 219, 8316, 25365, 8703, 233 https://tests.stockfishchess.org/tests/view/62e9a174400addce2c1346e4 closes https://github.com/official-stockfish/Stockfish/pull/4122 Bench: 5921315 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Stefan Geschwentner
Date: Sat Aug 6 13:50:01 2022 +0200 Timestamp: 1659786601 Tweak history updates In general the history update bonus is slightly decreased by 11% which gives a slower saturation speed. In addition only for main history the divisor is halfed (used history values are doubled to maintain same maximum) which have an effect in the opposite direction on saturation speed. STC: LLR: 2.95 (-2.94,2.94) <0.00,2.50> Total: 157088 W: 42673 L: 42168 D: 72247 Elo +1.12 Ptnml(0-2): 857, 17346, 41642, 17833, 866 https://tests.stockfishchess.org/tests/view/62e5517ab383a712b13867c5 LTC: LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 325592 W: 88705 L: 87753 D: 149134 Elo +1.02 Ptnml(0-2): 594, 32288, 96076, 33248, 590 https://tests.stockfishchess.org/tests/view/62e5e4f4b383a712b1387d53 closes https://github.com/official-stockfish/Stockfish/pull/4119 Bench: 5518728 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Sun Jul 31 11:03:28 2022 +0200 Timestamp: 1659258208 Do more TT cutoffs in case of exact bound The idea is that these TT entries are considered move valuable in TT replacement scheme - they are always overwriting other entries. So it makes sence for them to produce more aggressive cutoffs. passed STC https://tests.stockfishchess.org/tests/view/62e4d407b383a712b1385410 LLR: 2.95 (-2.94,2.94) <0.00,2.50> Total: 96632 W: 26045 L: 25659 D: 44928 Elo +1.39 Ptnml(0-2): 434, 10635, 25770, 11065, 412 passed LTC https://tests.stockfishchess.org/tests/view/62e523e2b383a712b1386193 LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 77960 W: 21363 L: 20989 D: 35608 Elo +1.67 Ptnml(0-2): 190, 7591, 23009, 8035, 155 closes https://github.com/official-stockfish/Stockfish/pull/4114 bench 5820568 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Dubslow
Date: Sun Jul 31 11:00:31 2022 +0200 Timestamp: 1659258031 remove useClassical depth condition passed STC: https://tests.stockfishchess.org/tests/view/62e0c3e98e4fa6ae472695ed LLR: 2.96 (-2.94,2.94) <-2.25,0.25> Total: 293568 W: 78934 L: 79151 D: 135483 Elo -0.26 Ptnml(0-2): 1344, 31488, 81366, 31213, 1373 passed LTC: https://tests.stockfishchess.org/tests/view/62e190aa8e4fa6ae4726b5b5 LLR: 2.98 (-2.94,2.94) <-2.25,0.25> Total: 187392 W: 50971 L: 51028 D: 85393 Elo -0.11 Ptnml(0-2): 384, 17801, 57369, 17772, 370 other attempts to otherwise tune this parameter failed, bounds 6,7,10,11 failed STC, 8 passed STC but failed LTC closes https://github.com/official-stockfish/Stockfish/pull/4112 bench 5796377 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Dubslow
Date: Sun Jul 24 09:18:38 2022 +0200 Timestamp: 1658647118 Simplify reduction condition for cutNodes LMR: for cutNodes, dont exclude killer moves. This was a prelude to reducing allNodes, altho that's failed so far. STC https://tests.stockfishchess.org/tests/view/62d64ad147ae1768b34a27c3 LLR: 2.95 (-2.94,2.94) <-2.25,0.25> Total: 37064 W: 10044 L: 9889 D: 17131 Elo +1.45 Ptnml(0-2): 162, 4115, 9828, 4260, 167 LTC https://tests.stockfishchess.org/tests/view/62d66cc047ae1768b34a2b14 LLR: 2.94 (-2.94,2.94) <-2.25,0.25> Total: 39832 W: 10796 L: 10659 D: 18377 Elo +1.19 Ptnml(0-2): 69, 3969, 11706, 4100, 72 closes https://github.com/official-stockfish/Stockfish/pull/4109 bench: 5697891 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Wed Jul 13 18:01:20 2022 +0200 Timestamp: 1657728080 Update default net to nn-ad9b42354671.nnue using trainer branch https://github.com/glinscott/nnue-pytorch/pull/208 with a slightly tweaked loss function (power 2.5 instead of 2.6), otherwise same training as in the previous net update https://github.com/official-stockfish/Stockfish/pull/4100 passed STC: LLR: 2.97 (-2.94,2.94) <0.00,2.50> Total: 367536 W: 99465 L: 98573 D: 169498 Elo +0.84 Ptnml(0-2): 1820, 40994, 97117, 42148, 1689 https://tests.stockfishchess.org/tests/view/62cc43fe50dcbecf5fc1c5b8 passed LTC: LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 25032 W: 6802 L: 6553 D: 11677 Elo +3.46 Ptnml(0-2): 40, 2424, 7341, 2669, 42 https://tests.stockfishchess.org/tests/view/62ce5f421dacb46e4d5fd277 closes https://github.com/official-stockfish/Stockfish/pull/4107 Bench: 5905619 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Wed Jul 13 18:00:39 2022 +0200 Timestamp: 1657728039 Simplify away some unneeded code in time management The lower bound of the clamp is never used since complexity can't be negative and thus is unneeded. closes https://github.com/official-stockfish/Stockfish/pull/4105 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Sat Jul 9 10:58:04 2022 +0200 Timestamp: 1657357084 Limit the researching at same depth. If the elapsed time is close to the available time, the time management thread can signal that the next iterations should be searched at the same depth (Threads.increaseDepth = false). While the rootDepth increases, the adjustedDepth is kept constant with the searchAgainCounter. In exceptional cases, when threading is used and the master thread, which controls the time management, signals to not increaseDepth, but by itself takes a long time to finish the iteration, the helper threads can search repeatedly at the same depth. This search finishes more and more quickly, leading to helper threads that report a rootDepth of MAX_DEPTH (245). The latter is not optimal as it is confusing for the user, stops search on these threads, and leads to an incorrect bias in the thread voting scheme. Probably with only a small impact on strength. This behavior was observed almost two years ago, see https://github.com/official-stockfish/Stockfish/issues/2717 This patch fixes #2717 by ensuring the effective depth increases at once every four iterations, even in increaseDepth is false. Depth 245 searches (for non-trivial positions) were indeed absent with this patch, but frequent with master in the tests below: https://discord.com/channels/435943710472011776/813919248455827515/994872720800088095 Total pgns: 2173 Base: 2867 Patch: 0 it passed non-regression testing in various setups: SMP STC: https://tests.stockfishchess.org/tests/view/62bfecc96178ffe6394ba036 LLR: 2.94 (-2.94,2.94) <-2.25,0.25> Total: 37288 W: 10171 L: 10029 D: 17088 Elo +1.32 Ptnml(0-2): 75, 3777, 10793, 3929, 70 SMP LTC: https://tests.stockfishchess.org/tests/view/62c08f6f49b62510394be066 LLR: 2.94 (-2.94,2.94) <-2.25,0.25> Total: 190568 W: 52125 L: 52186 D: 86257 Elo -0.11 Ptnml(0-2): 70, 17854, 59504, 17779, 77 LTC: https://tests.stockfishchess.org/tests/view/62c08b6049b62510394bdfb6 LLR: 2.96 (-2.94,2.94) <-2.25,0.25> Total: 48120 W: 13204 L: 13083 D: 21833 Elo +0.87 Ptnml(0-2): 54, 4458, 14919, 4571, 58 Special thanks to miguel-I, Disservin, ruicoelhopedro and others for analysing the problem, the data, and coming up with the key insight, needed to fix this longstanding issue. closes https://github.com/official-stockfish/Stockfish/pull/4104 Bench: 5182295 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Dubslow
Date: Sat Jul 9 10:53:17 2022 +0200 Timestamp: 1657356797 Time mgmt fix division. oversight changed the corresponding float division to integer division in a previous tune https://github.com/official-stockfish/Stockfish/commit/442c40b43de8ede1e424efa674c8d45322e3b43c it is stronger to keep the original float division. green LTC: https://tests.stockfishchess.org/tests/view/62bf34bc0340fb1e0cc934e7 LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 38952 W: 10738 L: 10467 D: 17747 Elo +2.42 Ptnml(0-2): 46, 3576, 11968, 3833, 53 yellow STC: https://tests.stockfishchess.org/tests/view/62bff6506178ffe6394ba1d1 LLR: -2.95 (-2.94,2.94) <0.00,2.50> Total: 226960 W: 61265 L: 61062 D: 104633 Elo +0.31 Ptnml(0-2): 938, 24398, 62582, 24647, 915 further slightly tweaked tests confirm this Elo gain. closes https://github.com/official-stockfish/Stockfish/pull/4097 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Tue Jul 5 13:55:50 2022 +0200 Timestamp: 1657022150 Simplify away FRC correction term Since new net is trained partially using FRC data this part of adjustment that penalises bishops that are locked in the corner is no longer needed - net should "know" this things itself much better. STC on FRC book : https://tests.stockfishchess.org/tests/view/62c3031b9e7d9997a12d852f LLR: 2.96 (-2.94,2.94) <-2.25,0.25> Total: 22048 W: 3003 L: 2845 D: 16200 Elo +2.49 Ptnml(0-2): 96, 1778, 7149, 1874, 127 LTC on FRC book : https://tests.stockfishchess.org/tests/view/62c32e939e7d9997a12d8c5e LLR: 2.94 (-2.94,2.94) <-2.25,0.25> Total: 36784 W: 3138 L: 3037 D: 30609 Elo +0.95 Ptnml(0-2): 36, 1842, 14537, 1939, 38 STC on DFRC book : https://tests.stockfishchess.org/tests/view/62c32efb9e7d9997a12d8c6f LLR: 2.94 (-2.94,2.94) <-2.25,0.25> Total: 20424 W: 3903 L: 3721 D: 12800 Elo +3.10 Ptnml(0-2): 172, 1984, 5724, 2154, 178 LTC on DFRC book : https://tests.stockfishchess.org/tests/view/62c358c79e7d9997a12d9319 LLR: 2.93 (-2.94,2.94) <-2.25,0.25> Total: 53784 W: 7581 L: 7480 D: 38723 Elo +0.65 Ptnml(0-2): 87, 3887, 18856, 3962, 100 closes https://github.com/official-stockfish/Stockfish/pull/4101 bench 5182295 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Mon Jul 4 15:42:34 2022 +0200 Timestamp: 1656942154 Update default net to nn-3c0054ea9860.nnu First things first... this PR is being made from court. Today, Tord and Stéphane, with broad support of the developer community are defending their complaint, filed in Munich, against ChessBase. With their products Houdini 6 and Fat Fritz 2, both Stockfish derivatives, ChessBase violated repeatedly the Stockfish GPLv3 license. Tord and Stéphane have terminated their license with ChessBase permanently. Today we have the opportunity to present our evidence to the judge and enforce that termination. To read up, have a look at our blog post https://stockfishchess.org/blog/2022/public-court-hearing-soon/ and https://stockfishchess.org/blog/2021/our-lawsuit-against-chessbase/ This PR introduces a net trained with an enhanced data set and a modified loss function in the trainer. A slight adjustment for the scaling was needed to get a pass on standard chess. passed STC: https://tests.stockfishchess.org/tests/view/62c0527a49b62510394bd610 LLR: 2.94 (-2.94,2.94) <0.00,2.50> Total: 135008 W: 36614 L: 36152 D: 62242 Elo +1.19 Ptnml(0-2): 640, 15184, 35407, 15620, 653 passed LTC: https://tests.stockfishchess.org/tests/view/62c17e459e7d9997a12d458e LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 28864 W: 8007 L: 7749 D: 13108 Elo +3.11 Ptnml(0-2): 47, 2810, 8466, 3056, 53 Local testing at a fixed 25k nodes resulted in Test run1026/easy_train_data/experiments/experiment_2/training/run_0/nn-epoch799.nnue localElo: 4.2 +- 1.6 The real strength of the net is in FRC and DFRC chess where it gains significantly. Tested at STC with slightly different scaling: FRC: https://tests.stockfishchess.org/tests/view/62c13a4002ba5d0a774d20d4 Elo: 29.78 +-3.4 (95%) LOS: 100.0% Total: 10000 W: 2007 L: 1152 D: 6841 Elo +29.78 Ptnml(0-2): 31, 686, 2804, 1355, 124 nElo: 59.24 +-6.9 (95%) PairsRatio: 2.06 DFRC: https://tests.stockfishchess.org/tests/view/62c13a5702ba5d0a774d20d9 Elo: 55.25 +-3.9 (95%) LOS: 100.0% Total: 10000 W: 2984 L: 1407 D: 5609 Elo +55.25 Ptnml(0-2): 51, 636, 2266, 1779, 268 nElo: 96.95 +-7.2 (95%) PairsRatio: 2.98 Tested at LTC with identical scaling: FRC: https://tests.stockfishchess.org/tests/view/62c26a3c9e7d9997a12d6caf Elo: 16.20 +-2.5 (95%) LOS: 100.0% Total: 10000 W: 1192 L: 726 D: 8082 Elo +16.20 Ptnml(0-2): 10, 403, 3727, 831, 29 nElo: 44.12 +-6.7 (95%) PairsRatio: 2.08 DFRC: https://tests.stockfishchess.org/tests/view/62c26a539e7d9997a12d6cb2 Elo: 40.94 +-3.0 (95%) LOS: 100.0% Total: 10000 W: 2215 L: 1042 D: 6743 Elo +40.94 Ptnml(0-2): 10, 410, 3053, 1451, 76 nElo: 92.77 +-6.9 (95%) PairsRatio: 3.64 This is due to the mixing in a significant fraction of DFRC training data in the final training round. The net is trained using the easy_train.py script in the following way: ``` python easy_train.py \ --training-dataset=../Leela-dfrc_n5000.binpack \ --experiment-name=2 \ --nnue-pytorch-branch=vondele/nnue-pytorch/lossScan4 \ --additional-training-arg=--param-index=2 \ --start-lambda=1.0 \ --end-lambda=0.75 \ --gamma=0.995 \ --lr=4.375e-4 \ --start-from-engine-test-net True \ --tui=False \ --seed=$RANDOM \ --max_epoch=800 \ --auto-exit-timeout-on-training-finished=900 \ --network-testing-threads 8 \ --num-workers 12 ``` where the data set used (Leela-dfrc_n5000.binpack) is a combination of our previous best data set (mix of Leela and some SF data) and DFRC data, interleaved to form: The data is available in https://drive.google.com/drive/folders/1S9-ZiQa_3ApmjBtl2e8SyHxj4zG4V8gG?usp=sharing Leela mix: https://drive.google.com/file/d/1JUkMhHSfgIYCjfDNKZUMYZt6L5I7Ra6G/view?usp=sharing DFRC: https://drive.google.com/file/d/17vDaff9LAsVo_1OfsgWAIYqJtqR8aHlm/view?usp=sharing The training branch used is https://github.com/vondele/nnue-pytorch/commits/lossScan4 A PR to the main trainer repo will be made later. This contains a revised loss function, now computing the loss from the score based on the win rate model, which is a more accurate representation than what we had before. Scaling constants are tweaked there as well. closes https://github.com/official-stockfish/Stockfish/pull/4100 Bench: 5186781 see source |