0716b845fd
Update NNUE architecture to SFNNv9 and net nn-ae6a388e4a1a.nnue
...
Part 1: PyTorch Training, linrock
Trained with a 10-stage sequence from scratch, starting in May 2023:
https://github.com/linrock/nnue-tools/blob/master/exp-sequences/3072-10stage-SFNNv9.yml
While the training methods were similar to the L1-2560 training sequence,
the last two stages introduced min-v2 binpacks,
where bestmove capture and in-check position scores were not zeroed during minimization,
for compatibility with skipping SEE >= 0 positions and future research.
Training data can be found at:
https://robotmoon.com/nnue-training-data
This net was tested at epoch 679 of the 10th training stage:
https://tests.stockfishchess.org/tests/view/65f32e460ec64f0526c48dbc
Part 2: SPSA Training, Viren6
The net was then SPSA tuned.
This consisted of the output weights (32 * 8) and biases (8)
as well as the L3 biases (32 * 8) and L2 biases (16 * 8), totalling 648 params in total.
The SPSA tune can be found here:
https://tests.stockfishchess.org/tests/view/65fc33ba0ec64f0526c512e3
With the help of Disservin , the initial weights were extracted with:
https://github.com/Viren6/Stockfish/tree/new228
The net was saved with the tuned weights using:
https://github.com/Viren6/Stockfish/tree/new241
Earlier nets of the SPSA failed STC compared to the base 3072 net of part 1:
https://tests.stockfishchess.org/tests/view/65ff356e0ec64f0526c53c98
Therefore it is suspected that the SPSA at VVLTC has
added extra scaling on top of the scaling of increasing the L1 size.
Passed VVLTC 1:
https://tests.stockfishchess.org/tests/view/6604a9020ec64f0526c583da
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 53042 W: 13554 L: 13256 D: 26232
Ptnml(0-2): 12, 5147, 15903, 5449, 10
Passed VVLTC 2:
https://tests.stockfishchess.org/tests/view/660ad1b60ec64f0526c5dd23
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 17506 W: 4574 L: 4315 D: 8617
Ptnml(0-2): 1, 1567, 5362, 1818, 5
STC Elo estimate:
https://tests.stockfishchess.org/tests/view/660b834d01aaec5069f87cb0
Elo: -7.66 ± 3.8 (95%) LOS: 0.0%
Total: 9618 W: 2440 L: 2652 D: 4526
Ptnml(0-2): 80, 1281, 2261, 1145, 42
nElo: -13.94 ± 6.9 (95%) PairsRatio: 0.87
closes https://tests.stockfishchess.org/tests/view/660b834d01aaec5069f87cb0
bench 1823302
Co-Authored-By: Linmiao Xu <lin@robotmoon.com >
2024-04-02 08:49:48 +02:00
Viren6 and Joost VandeVondele
f2b6b5cfc9
Introduce Triple Extensions
...
This replaces singularquietLMR with triple instead of double extending non-capture ttmoves that have value far below singularBeta. This threshold value is initially set to 200, there is scope for more scaling by reducing it as occured with double extensions.
Passed STC:
https://tests.stockfishchess.org/tests/view/65b683b8c865510db0274074
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 222912 W: 58141 L: 57535 D: 107236
Ptnml(0-2): 1063, 26244, 56154, 27014, 981
Passed LTC:
https://tests.stockfishchess.org/tests/view/65bae6d4c865510db0278eb5
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 66306 W: 16825 L: 16440 D: 33041
Ptnml(0-2): 40, 7374, 17952, 7735, 52
closes https://github.com/official-stockfish/Stockfish/pull/5027
bench 1394701
2024-02-03 17:26:42 +01:00
Viren6 and Disservin
fcbb02ffde
Use ttPv in depth condition of singular extensions
...
This replaces the PvNode condition and tte Pv call previously with using
the precomputed ttPv, and also removes the multiplier of 2. This new
depth condition occurs with approximately equal frequency (47%) to the
old depth condition (measured when the other conditions in the if are
true), so non-linear scaling behaviour isn't expected.
Passed Non-Reg STC:
https://tests.stockfishchess.org/tests/view/65b0e132c865510db026da27
LLR: 2.97 (-2.94,2.94) <-1.75,0.25>
Total: 243232 W: 62432 L: 62437 D: 118363
Ptnml(0-2): 910, 28937, 61900, 28986, 883
Passed Non-Reg LTC:
https://tests.stockfishchess.org/tests/view/65b2053bc865510db026eea1
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 190596 W: 47666 L: 47618 D: 95312
Ptnml(0-2): 115, 21710, 51596, 21766, 111
closes https://github.com/official-stockfish/Stockfish/pull/5015
Bench: 1492957
2024-01-26 21:15:07 +01:00
Viren6 and Disservin
aa15a9179b
Refactor ttPv reduction conditions
...
closes https://github.com/official-stockfish/Stockfish/pull/4999
No functional change
2024-01-21 12:33:08 +01:00
Viren6 and Disservin
c8bc2ce4fa
Improve ttPv reduction
...
This patch allows a partial reduction decrease when a node is likely to
fail low, and increases the reduction decrease when a node has failed
high.
Passed STC:
https://tests.stockfishchess.org/tests/view/65a626e779aa8af82b9722bc
LLR: 2.93 (-2.94,2.94) <0.00,2.00>
Total: 157824 W: 40332 L: 39835 D: 77657
Ptnml(0-2): 543, 18617, 40098, 19108, 546
Passed LTC:
https://tests.stockfishchess.org/tests/view/65a7290279aa8af82b97328a
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 57228 W: 14475 L: 14111 D: 28642
Ptnml(0-2): 34, 6278, 15633, 6628, 41
closes https://github.com/official-stockfish/Stockfish/pull/4994
Bench: 1364759
2024-01-17 18:56:37 +01:00
Viren6 and Disservin
28f8663f39
Modify ttPV reduction
...
This patch modifies ttPV reduction by reducing 1 more unless ttValue is above alpha.
Inspired from @pb00068 https://tests.stockfishchess.org/tests/view/658060796a3b4f6202215f1f
Passed STC:
https://tests.stockfishchess.org/tests/view/6591867679aa8af82b958328
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 37856 W: 9727 L: 9407 D: 18722
Ptnml(0-2): 99, 4444, 9568, 4672, 145
Passed LTC:
https://tests.stockfishchess.org/tests/view/6591d9b679aa8af82b958a6c
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 128256 W: 32152 L: 31639 D: 64465
Ptnml(0-2): 64, 14364, 34772, 14851, 77
closes https://github.com/official-stockfish/Stockfish/pull/4957
Bench: 1176235
2024-01-04 15:49:33 +01:00
Viren6 and Joost VandeVondele
887bbd8b3d
Remove setting of static to none if in check in qsearch
...
Small simplification
Passed non-regression STC:
https://tests.stockfishchess.org/tests/view/6487924d713491385c8034ae
LLR: 2.93 (-2.94,2.94) <-1.75,0.25>
Total: 59616 W: 15885 L: 15703 D: 28028
Ptnml(0-2): 144, 6130, 17086, 6296, 152
closes https://github.com/official-stockfish/Stockfish/pull/4618
No functional change.
2023-06-14 18:33:56 +02:00
