Active : 3 Machines / 69 Threads / 87.3 MNPS
Priority 0
DsekerBlackMarlinnmp-eval-increase-3diff40.0+0.40LLR: 1.65 (-2.94, 2.94) [0.00, 4.00]
Games: 2880 W: 731 L: 652 D: 1497
Ptnml(0-2): 3, 299, 760, 372, 6
early stop NMP if eval after null move is better than eval before (take 3)
DsekerStashno_goto_when_in_checkdiff40.0+0.40LLR: 1.57 (-2.94, 2.94) [-4.00, 1.00]
Games: 16020 W: 3734 L: 3713 D: 8573
Ptnml(0-2): 76, 1887, 4074, 1886, 87
Don't use a goto to reach the main search loop when in check
ConnorSeernon-pawn-correction-historydiff40.0+0.40LLR: 0.82 (-2.94, 2.94) [0.00, 3.00]
Games: 52904 W: 12924 L: 12765 D: 27215
Ptnml(0-2): 78, 6006, 14131, 6153, 84
test
VshcheIgelshared_hist2diff5.0+0.05LLR: 0.23 (-2.94, 2.94) [0.00, 2.00]
Games: 716 W: 184 L: 163 D: 369
Ptnml(0-2): 1, 73, 189, 94, 1
3.8.11: shared history
Priority -1
VshcheIgelshared_hist2diff20.0+0.20LLR: 0.00 (-2.94, 2.94) [0.00, 2.00]
Games: 0 W: 0 L: 0 D: 0
Ptnml(0-2): 0, 0, 0, 0, 0
3.8.11: shared history
DsekerBlackMarlinnmp-eval-increase-2diff8.0+0.08LLR: -1.39 (-2.94, 2.94) [0.00, 4.00]
Games: 8752 W: 2144 L: 2182 D: 4426
Ptnml(0-2): 55, 1066, 2172, 1028, 55
early stop NMP if eval after null move is better than eval before (take 2)
DsekerBlackMarlinmovegen-depth-seediff8.0+0.08LLR: -1.41 (-2.94, 2.94) [0.00, 4.00]
Games: 20038 W: 5002 L: 5006 D: 10030
Ptnml(0-2): 131, 2427, 4904, 2429, 128
depth based see margin for good captures in move gen
DsekerBlackMarlinlmr-combineddiff8.0+0.08LLR: -1.53 (-2.94, 2.94) [0.00, 4.00]
Games: 43980 W: 10993 L: 10928 D: 22059
Ptnml(0-2): 293, 5206, 10959, 5207, 325
Reduce more when all 3 reduction conditions are met
DsekerBlackMarlinsimplify-killer-lmrdiff40.0+0.40LLR: -1.80 (-2.94, 2.94) [-4.00, 0.00]
Games: 46366 W: 11168 L: 11378 D: 23820
Ptnml(0-2): 103, 5417, 12334, 5245, 84
Simplify killer LMR
DsekerBlackMarlinnmp-eval-increasediff8.0+0.08LLR: -1.91 (-2.94, 2.94) [0.00, 4.00]
Games: 18270 W: 4449 L: 4482 D: 9339
Ptnml(0-2): 120, 2203, 4532, 2150, 130
early stop NMP if eval after null move is better than eval before
DsekerBlackMarlinmovegen-depth-capture-valuediff8.0+0.08LLR: -2.11 (-2.94, 2.94) [0.00, 4.00]
Games: 13624 W: 3381 L: 3437 D: 6806
Ptnml(0-2): 94, 1592, 3487, 1554, 85
Disregard the value of captured pieces in movegen at higher depths
Priority -5
TerjeWeissConnectivity2diff60.0+0.60LLR: -0.21 (-2.94, 2.94) [0.00, 3.00]
Games: 68704 W: 17741 L: 17594 D: 33369
Ptnml(0-2): 461, 8109, 17058, 8270, 454
Finished
VshcheIgelnn-epoch999 *diff60.0+0.60LLR: 2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 3392 W: 885 L: 614 D: 1893
Ptnml(0-2): 4, 293, 838, 550, 11
test against frc net
DsekerBlackMarlinnmp-eval-increase-3diff8.0+0.08LLR: 2.95 (-2.94, 2.94) [0.00, 4.00]
Games: 101338 W: 25264 L: 24811 D: 51263
Ptnml(0-2): 642, 11874, 25263, 12169, 721
early stop NMP if eval after null move is better than eval before (take 3)
VshcheIgelnumadiff90.0+0.90LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 44568 W: 11399 L: 11089 D: 22080
Ptnml(0-2): 16, 4999, 11940, 5317, 12
numa + new net - VLTC - TBC
MhouppStashno_goto_when_in_checkdiff8.0+0.08LLR: 3.07 (-2.94, 2.94) [-4.00, 1.00]
Games: 5800 W: 1545 L: 1432 D: 2823
Ptnml(0-2): 80, 652, 1334, 743, 91
Don't use a goto to reach the main search loop when in check
VshcheIgelshared_histdiff5.0+0.05LLR: -0.21 (-2.94, 2.94) [0.00, 2.00]
Games: 4942 W: 1237 L: 1248 D: 2457
Ptnml(0-2): 4, 589, 1293, 584, 1
3.8.10: shared history
DsekerBlackMarlinquiet-bad-cap-lmr-2diff8.0+0.08LLR: -3.05 (-2.94, 2.94) [0.00, 4.00]
Games: 4676 W: 1097 L: 1222 D: 2357
Ptnml(0-2): 25, 611, 1185, 498, 19
More LMR for quiet moves and bad captures excluding killers
VshcheIgelhist_capdiff10.0+0.10LLR: -1.08 (-2.94, 2.94) [-3.00, 1.00]
Games: 34894 W: 8906 L: 9003 D: 16985
Ptnml(0-2): 69, 4118, 9182, 3997, 81
3.8.9: cap history bonus
RosentWinterwider_hidden *diff8.0+0.08LLR: -2.97 (-2.94, 2.94) [0.00, 5.00]
Games: 6878 W: 1700 L: 1804 D: 3374
Ptnml(0-2): 180, 862, 1445, 786, 166
This network has a larger hidden layer. This is unlikely to be worth the overhead at this point.
VshcheIgelhist_capdiff60.0+0.60LLR: 2.96 (-2.94, 2.94) [-3.00, 1.00]
Games: 27862 W: 7027 L: 6942 D: 13893
Ptnml(0-2): 7, 3124, 7590, 3197, 13
3.8.9: cap history bonus
VshcheIgelB61B92877D7AD253diff1.0+0.01LLR: 2.97 (-2.94, 2.94) [-3.00, 1.00]
Games: 6230 W: 1735 L: 1613 D: 2882
Ptnml(0-2): 15, 681, 1605, 795, 19
3.8.8: calculate nodes share per move
ConnorSeernon-pawn-correction-historydiff8.0+0.08LLR: 2.95 (-2.94, 2.94) [0.00, 3.00]
Games: 25588 W: 6411 L: 6170 D: 13007
Ptnml(0-2): 150, 2905, 6452, 3128, 159
test
VshcheIgelnodes_sharediff60.0+0.60LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 17598 W: 4622 L: 4351 D: 8625
Ptnml(0-2): 7, 1874, 4772, 2133, 13
3.8.8: calculate nodes share per move
JhonnoBerserkexp/h640diff60.0+0.60LLR: -2.25 (-2.25, 2.89) [0.00, 2.50]
Games: 7762 W: 1894 L: 2032 D: 3836
Ptnml(0-2): 12, 936, 2116, 812, 5
0052: N_HIDDEN 640 (main recipe, two stages) vs main net; 128-lane accumulator tiles. Fixed 40k nodes +6.73 +/- 3.23, SPRT [0, 2.5] passed (16,471 games)
VshcheIgelnodes_sharediff10.0+0.10Elo: 45.85 +- 6.19 (95%) [N=40000]
Games: 3308 W: 1082 L: 648 D: 1578
Ptnml(0-2): 10, 227, 753, 647, 17
Regression peek
VshcheIgelnodes_sharediff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 18420 W: 4996 L: 4713 D: 8711
Ptnml(0-2): 32, 2017, 4843, 2272, 46
3.8.8: calculate nodes share per move
RosentWintergraph_history_hack *diff8.0+0.08LLR: 2.99 (-2.94, 2.94) [-5.00, 0.00]
Games: 75762 W: 20097 L: 20307 D: 35358
Ptnml(0-2): 1415, 9166, 16865, 9084, 1351
Cleaner version and testing without win adjudication.
VshcheIgelnn-epoch999 *diff10.0+0.10LLR: 2.97 (-2.94, 2.94) [0.00, 2.00]
Games: 5662 W: 1577 L: 1268 D: 2817
Ptnml(0-2): 56, 561, 1283, 880, 51
test against frc net
JhonnoBerserkfeat/lmr-ltc-tunediff60.0+0.60LLR: -1.93 (-2.25, 2.89) [0.00, 2.50]
Games: 43624 W: 10515 L: 10569 D: 22540
Ptnml(0-2): 18, 5096, 11642, 5034, 22
LMR: LTC-tuned parameters
JhonnoBerserkspsa-lmr-ltcdiff60.0+0.60Tuning 11 Parameters
11482/25000 Iterations
183720/400000 Games Played
LTC SPSA: LMR (pgns)
VshcheIgelnumadiff60.0+0.60LLR: 2.97 (-2.94, 2.94) [0.00, 2.00]
Games: 23646 W: 6109 L: 5825 D: 11712
Ptnml(0-2): 12, 2606, 6306, 2884, 15
numa + new net - LTC
JhonnoBerserkfeat/lmr-capturediff10.0+0.10LLR: -2.29 (-2.25, 2.89) [0.00, 2.00]
Games: 45638 W: 11198 L: 11334 D: 23106
Ptnml(0-2): 120, 5525, 11665, 5389, 120
LMR: reduce captures one ply less
JhonnoBerserkfeat/lmr-lower-basediff10.0+0.10LLR: -2.28 (-2.25, 2.89) [0.00, 2.00]
Games: 13170 W: 3140 L: 3322 D: 6708
Ptnml(0-2): 40, 1654, 3361, 1508, 22
LMR: lower LMR_BASE by ~1/3 ply
JhonnoBerserkfeat/lmr-ttpv-fail-lowdiff10.0+0.10LLR: -2.30 (-2.94, 2.94) [0.00, 2.00]
Games: 29546 W: 7207 L: 7369 D: 14970
Ptnml(0-2): 81, 3686, 7393, 3540, 73
LMR: +1 ply on ttPv nodes whose TT score fails low
JhonnoBerserkfeat/lmr-history-continuousdiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [-2.00, 0.00]
Games: 37076 W: 9262 L: 9055 D: 18759
Ptnml(0-2): 79, 4344, 9485, 4551, 79
LMR: continuous history term (LMR_HISTORY 6144), LMR_BASE -1/16 ply; non-regression
JhonnoBerserk0051-pawnpairs-h512-s2see-sorted.nndiff60.0+0.60LLR: 0.65 (-2.25, 2.89) [0.00, 2.50]
Games: 20086 W: 4759 L: 4681 D: 10646
Ptnml(0-2): 5, 2248, 5463, 2318, 9
0051: 0050 with Plenty inspired stage-2 SEE filter