69bc3ef9be
Output loss more often.
2020-11-14 12:33:25 +09:00
a71623f74c
Add explicit read head seek to the start of the binpack file. Otherwise on MACOS the read head is placed at the end when app is specified.
2020-11-13 19:56:36 +09:00
285bf7041a
Increase reduction at root
...
when the best move does not change frequently
STC:
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 51320 W: 5159 L: 4956 D: 41205
Ptnml(0-2): 215, 3897, 17242, 4082, 224
https://tests.stockfishchess.org/tests/view/5faa072367cbf42301d6a767
LTC:
LLR: 2.98 (-2.94,2.94) {0.25,1.25}
Total: 15952 W: 762 L: 642 D: 14548
Ptnml(0-2): 8, 561, 6725, 667, 15
https://tests.stockfishchess.org/tests/view/5faa4c3567cbf42301d6a794
closes https://github.com/official-stockfish/Stockfish/pull/3225
Bench: 3954692
2020-11-12 12:49:03 +01:00
b5781150ea
Increase reduction based on the number of best move changes.
...
Thanks to Vizvezdenec for the PvNode idea and also to vondele the !PvNode idea.
Passed STC:
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 19120 W: 1998 L: 1839 D: 15283
Ptnml(0-2): 76, 1445, 6375, 1572, 92
https://tests.stockfishchess.org/tests/view/5fa8af3e67cbf42301d6a6c9
Passed LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 75584 W: 3454 L: 3205 D: 68925
Ptnml(0-2): 54, 2832, 31771, 3081, 54
closes https://github.com/official-stockfish/Stockfish/pull/3224
Bench: 3595418
2020-11-10 18:21:05 +01:00
2a8576b804
Fix compilation issues.
2020-11-10 10:21:09 +09:00
8069963c56
Update convert docs.
2020-11-10 10:21:09 +09:00
5d88e7bce8
Add optional move validation to training data conversion. No longer rely on static initialization order for magics initialization.
2020-11-10 10:21:09 +09:00
392b529c3f
Qsearch pruning: follow-up
...
This is a follow-up of the recent qsearch pruning patch in
https://github.com/official-stockfish/Stockfish/commit/a260c9a8a24a2630a900efc3821000c3481b0c5d
We now use the same guard condition (testing that we already have a defense with
a score better score than a TB loss) for all pruning heuristics in qsearch().
This allows some pruning when in check, but in a controlled way to ensure that
no wrong mate scores appear.
Tested with Elo-gaining bounds:
STC:
LLR: 2.97 (-2.94,2.94) {-0.25,1.25}
Total: 22632 W: 2433 L: 2264 D: 17935
Ptnml(0-2): 98, 1744, 7487, 1865, 122
https://tests.stockfishchess.org/tests/view/5fa59405936c54e11ec99515
LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 105432 W: 4965 L: 4648 D: 95819
Ptnml(0-2): 85, 4110, 44011, 4423, 87
https://tests.stockfishchess.org/tests/view/5fa5b609936c54e11ec9952a
closes https://github.com/official-stockfish/Stockfish/pull/3221
Bench: 3578092
2020-11-08 09:15:34 +01:00
32edb1d009
Update default net to nn-c3ca321c51c9.nnue
...
Optimization of the net biases of the 32 x 32 layer and the output layer.
Tuning of 32 x 32 layer (200k games, 5 seconds TC)
https://tests.stockfishchess.org/tests/view/5f9aaf266a2c112b60691c68
STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 41848 W: 4665 L: 4461 D: 32722
Ptnml(0-2): 239, 3308, 13659, 3446, 272
https://tests.stockfishchess.org/tests/view/5fa5ef5a936c54e11ec9954f
LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 88008 W: 4045 L: 3768 D: 80195
Ptnml(0-2): 69, 3339, 36908, 3622, 66
https://tests.stockfishchess.org/tests/view/5fa62a78936c54e11ec99577
closes https://github.com/official-stockfish/Stockfish/pull/3220
Bench: 3649288
2020-11-08 08:36:16 +01:00
ba35c88ab8
AVX-512 for smaller affine and feature transforms.
...
For the feature transformer the code is analogical to AVX2 since there was room for easy adaptation of wider simd registers.
For the smaller affine transforms that have 32 byte stride we keep 2 columns in one zmm register. We also unroll more aggressively so that in the end we have to do 16 parallel horizontal additions on ymm slices each consisting of 4 32-bit integers. The slices are embedded in 8 zmm registers.
These changes provide about 1.5% speedup for AVX-512 builds.
Closes https://github.com/official-stockfish/Stockfish/pull/3218
No functional change.
2020-11-07 16:49:49 +01:00
7fc47eeb6f
Introducing King On File
...
this new concept calculates bonuses/penalties for the king when the king is in a semiopen or open file.
Passed STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 44904 W: 9365 L: 9028 D: 26511
Ptnml(0-2): 857, 5309, 9841, 5530, 915
https://tests.stockfishchess.org/tests/view/5fa343625d72639a7acef72b
Passed LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 60552 W: 8449 L: 8051 D: 44052
Ptnml(0-2): 466, 5772, 17481, 6012, 545
https://tests.stockfishchess.org/tests/view/5fa40e365d72639a7acef79e
closes https://github.com/official-stockfish/Stockfish/pull/3219
Bench: 3689484
2020-11-07 16:49:49 +01:00
04a320666e
Change handling the special case of a single legal move.
...
Using no searching time in case of a single legal move is not beneficial from
a strength point of view, and this special case can be easily removed:
STC:
LLR: 2.93 (-2.94,2.94) {-1.25,0.25}
Total: 22472 W: 2458 L: 2357 D: 17657
Ptnml(0-2): 106, 1733, 7453, 1842, 102
https://tests.stockfishchess.org/tests/view/5f926cbc81eda81bd78cb6df
LTC:
LLR: 2.94 (-2.94,2.94) {-0.75,0.25}
Total: 37880 W: 1736 L: 1682 D: 34462
Ptnml(0-2): 22, 1392, 16057, 1448, 21
https://tests.stockfishchess.org/tests/view/5f92a26081eda81bd78cb6fe
The advantage of using the normal time management for a single legal move is that scores
reported for that move are reasonable, not searching leads to artifacts during games
(see e.g. https://tcec-chess.com/#div=sf&game=96&season=19 )
The disadvantage of using normal time management of a single legal move is that thinking
times can be unnaturally long, making it 'painful to watch' in online tournaments.
This patch uses normal time management, but caps the used time to 500ms.
This should lead to reasonable scores, and be hardly perceptible.
closes https://github.com/official-stockfish/Stockfish/pull/3195
closes https://github.com/official-stockfish/Stockfish/pull/3183
variant of a patch suggested by SFisGOD
No functional change.
2020-11-07 16:48:02 +01:00
3f6451eff7
Manually align arrays on the stack
...
as a workaround to issues with overaligned alignas() on stack variables in gcc < 9.3 on windows.
closes https://github.com/official-stockfish/Stockfish/pull/3217
fixes #3216
No functional change
2020-11-04 19:52:42 +01:00
a260c9a8a2
Fix incorrect pruning in qsearch
...
Only do countermove based pruning in qsearch if we already have a move with a better score than a TB loss.
This patch fixes a bug (started as 843a961#3171 and later in #3199 , #3198 and #3210 .
This PR effectively closes #3171
It also likely fixes #3196 where this causes user visible incorrect TB scores,
which probably result from these incorrect mate scores.
Passed STC and LTC non-regression tests.
https://tests.stockfishchess.org/tests/view/5f9ef8dabca9bf35bae7f648
LLR: 2.93 (-2.94,2.94) {-1.25,0.25}
Total: 21672 W: 2339 L: 2230 D: 17103
Ptnml(0-2): 126, 1689, 7083, 1826, 112
https://tests.stockfishchess.org/tests/view/5f9f0caebca9bf35bae7f666
LLR: 2.97 (-2.94,2.94) {-0.75,0.25}
Total: 33152 W: 1551 L: 1485 D: 30116
Ptnml(0-2): 27, 1308, 13832, 1390, 19
closes https://github.com/official-stockfish/Stockfish/pull/3214
Bench: 3625915
2020-11-02 19:41:17 +01:00
931070b65a
Elo Worth in King Danger
...
Adding the EloWorth for each term in King Danger.
Should be useful for simplifications, tuning patches, and new ideas.
closes https://github.com/official-stockfish/Stockfish/pull/3204
non-functional change
2020-11-02 19:41:17 +01:00
75e06a1c89
Optimize affine transform for SSSE3 and higher targets.
...
A non-functional speedup. Unroll the loops going over
the output dimensions in the affine transform layers by
a factor of 4 and perform 4 horizontal additions at a time.
Instead of doing naive horizontal additions on each vector
separately use hadd and shuffling between vectors to reduce
the number of instructions by using all lanes for all stages
of the horizontal adds.
passed STC of the initial version:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 17808 W: 1914 L: 1756 D: 14138
Ptnml(0-2): 76, 1330, 5948, 1460, 90
https://tests.stockfishchess.org/tests/view/5f9d516f6a2c112b60691da3
passed STC of the final version after cleanup:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 16296 W: 1750 L: 1595 D: 12951
Ptnml(0-2): 72, 1192, 5479, 1319, 86
https://tests.stockfishchess.org/tests/view/5f9df5776a2c112b60691de3
closes https://github.com/official-stockfish/Stockfish/pull/3203
No functional change
2020-11-02 19:41:17 +01:00
dfc7f88650
Update default net to nn-cb26f10b1fd9.nnue
...
Result of https://tests.stockfishchess.org/tests/view/5f9a06796a2c112b60691c0f tuning.
STC
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 53712 W: 5776 L: 5561 D: 42375
Ptnml(0-2): 253, 4282, 17604, 4431, 286
https://tests.stockfishchess.org/tests/view/5f9c7bbc6a2c112b60691d4d
LTC
LLR: 2.97 (-2.94,2.94) {0.25,1.25}
Total: 80184 W: 4007 L: 3739 D: 72438
Ptnml(0-2): 58, 3302, 33130, 3518, 84
https://tests.stockfishchess.org/tests/view/5f9d01f06a2c112b60691d87
closes https://github.com/official-stockfish/Stockfish/pull/3209
bench: 3517795
2020-11-01 08:02:40 +01:00
987b6c98d4
Move the observed feature collection to the threaded part now that it can be done safely.
2020-11-01 11:02:44 +09:00
c53be1b23f
Add specialized bitset for use in the trainer for observed features tracking.
2020-11-01 11:02:44 +09:00
e8907bcfc4
Replace omp in trainer_feature_transformer
2020-10-31 11:54:03 +09:00
db1b33d4ac
Optimize trainer clipped relu propagate
2020-10-31 11:52:51 +09:00
b5714c4084
Parallelize input slice trainer backprop.
2020-10-31 11:52:26 +09:00
941897ff2c
Optimize trainer clipped relu backpropagate.
2020-10-31 11:50:12 +09:00
c96743c5bd
Optimize feature transformer backpropagation stats.
2020-10-31 11:49:29 +09:00
2c10b1babc
Optimize feature transformer clipped relu.
2020-10-31 11:48:02 +09:00
7bedf6c5ab
Specify the whole evalsave message because otherwise the first evalsave/0 triggers it.
2020-10-31 08:36:58 +09:00
8c81bbd3db
Fix the counter in for_each_index_with_workers going out of scope before workers finish.
2020-10-31 08:36:58 +09:00
a56d8124d8
Replace non-blas parts of trainers with our own blas-like routines.
2020-10-31 08:36:58 +09:00
c56a4a36eb
Add our own blas-like routines that use stockfish's thread pool for parallelization.
2020-10-29 23:57:51 +09:00
ee0917a345
Pass ThreadPool to update_parameters, propagate, and backpropagate.
2020-10-29 09:21:19 +09:00
f1e96cab55
Align trainer arrays to cache line.
2020-10-29 09:12:50 +09:00
8fac468259
Add a cache line aligned allocator.
2020-10-29 09:12:50 +09:00
ec9e49e875
Add a HalfKA architecture (a product of K - king, and A - any piece) along with all required infrastructure. HalfKA doesn't discriminate kings compared to HalfKP. Keep old architecture as the default one.
2020-10-29 09:10:01 +09:00
0f6c08c73f
Do not skip non-recapture ttMove when in check
...
The qsearch() MovePicker incorrectly skips a non-recapture ttMove
when in check (if depth <= DEPTH_QS_RECAPTURES). This is clearly not
intended and can cause qsearch() to return a mate score when there
is no mate. Introduced in cad300c6596f0e#3171 and #3198 .
This PR fixes the bug by not skipping the non-recapture ttMove when in check.
Passed non-regression STC:
https://tests.stockfishchess.org/tests/view/5f9867ea6a2c112b60691b10
LLR: 2.98 (-2.94,2.94) {-1.25,0.25}
Total: 27112 W: 2943 L: 2842 D: 21327
Ptnml(0-2): 127, 2170, 8878, 2237, 144
Passed non-regression LTC:
https://tests.stockfishchess.org/tests/view/5f9967326a2c112b60691bb0
LLR: 2.99 (-2.94,2.94) {-0.75,0.25}
Total: 18392 W: 807 L: 738 D: 16847
Ptnml(0-2): 9, 655, 7802, 718, 12
closes https://github.com/official-stockfish/Stockfish/pull/3199
closes https://github.com/official-stockfish/Stockfish/pull/3198
Bench: 3870606
2020-10-28 23:40:37 +01:00
317fda2516
Cleanup eval saving and lr scheduling.
2020-10-28 23:08:05 +09:00
6328135264
Update default net to nn-2eb2e0707c2b.nnue
...
Optimization of the net weights of the 32 x 32 layer (1024 parameters) and net biases of the 512 x 32 layer (32 parameters) using SPSA.
Tuning of 32 x 32 Layer (800,000 games, 5 seconds time control):
https://tests.stockfishchess.org/tests/view/5f942040d3978d7e86f1aa05
Tuning of 512 x 32 Layer (80,000 games, 20 seconds time control):
https://tests.stockfishchess.org/tests/view/5f8f926d2c92c7fe3a8c608b
STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 17336 W: 1918 L: 1754 D: 13664
Ptnml(0-2): 79, 1344, 5672, 1480, 93
https://tests.stockfishchess.org/tests/view/5f9882346a2c112b60691b34
LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 37304 W: 1822 L: 1651 D: 33831
Ptnml(0-2): 27, 1461, 15501, 1640, 23
https://tests.stockfishchess.org/tests/view/5f98a4b36a2c112b60691b40
closes https://github.com/official-stockfish/Stockfish/pull/3201
Bench: 3403528
2020-10-28 08:13:34 +01:00
bde3505758
Bishop Pawns based on Files
...
Passed STC:
https://tests.stockfishchess.org/tests/view/5f8cc8145a4eacb45305da3c
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 132544 W: 27795 L: 27328 D: 77421
Ptnml(0-2): 2756, 15558, 29272, 15835, 2851
Passed LTC:
https://tests.stockfishchess.org/tests/view/5f8df614bacb75a4f9a4721e
LLR: 2.94 (-2.94,2.94) {0.25,1.25}
Total: 169608 W: 23257 L: 22622 D: 123729
Ptnml(0-2): 1408, 16316, 48758, 16877, 1445
closes https://github.com/official-stockfish/Stockfish/pull/3194
Bench: 4067106
2020-10-28 08:11:29 +01:00
680654b254
Add dots to output every epoch for progress visualization.
2020-10-28 09:36:43 +09:00
f81fa3d712
Replace global_learning_rate with learning_rate local to the learner and passed to update_parameters as a parameter.
2020-10-28 09:36:07 +09:00
cde6ec2bf2
Make all grad related functions in learn static. Pass calc_grad as a parameter.
2020-10-27 14:47:50 +09:00
e4868cb59e
Move setting learn search limits to learner.
2020-10-27 14:47:07 +09:00
c229929d26
Remove the position parameter from learn.
2020-10-27 00:35:43 +09:00
a8066cd4a9
Rename elmo lambdas
2020-10-27 00:33:58 +09:00
f7de49eb66
Create a collective parameter struct for learner.
2020-10-27 00:33:58 +09:00
ba390a7f9a
Print the used factorizer when intializing training.
2020-10-27 00:32:39 +09:00
e01397c674
Remove multi_think
2020-10-26 19:40:40 +09:00
e515f1f61f
Move SfenWriter to a separate file
2020-10-26 19:39:58 +09:00
65e443954a
Update expected gensfen finished responses.
2020-10-26 09:37:59 +09:00
03abfae41f
Reorder members, renaming.
2020-10-26 09:37:59 +09:00
6d4d20c4be
Cleaner printing and some renaming.
2020-10-26 09:37:59 +09:00
Tomasz Sobczyk