Stockfish Testing Queue

Finished - 27247 tests

09-04-15 Vo Swordfish diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 26446 W: 5092 L: 4980 D: 16374
sprt @ 15+0.05 th 1 Sync up lmr and move ordering. This time test as simplification.
09-04-15 aj hanging_A diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7012 W: 1329 L: 1416 D: 4267
sprt @ 15+0.05 th 1 Give additional hanging pawn bonus only for bigger piece attacking smaller piece: STC
09-04-15 sg update_stats_pv2 diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 41485 W: 8047 L: 7830 D: 25608
sprt @ 15+0.05 th 1 update stats also for non-cutoff best move at pv nodes. Test against my passed 'update stats also in check' patch (Take 2)
09-04-15 sg update_stats_pv diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 21697 W: 4279 L: 4114 D: 13304
sprt @ 15+0.05 th 1 update stats also for non-cutoff best move at pv nodes. Test against my passed 'update stats also in check' patch
07-04-15 Fi mPickScore diff
LLR: 3.51 (-2.94,2.94) [0.00,4.00]
Total: 125092 W: 20032 L: 19468 D: 85592
sprt @ 60+0.05 th 1 MovePicker::score<QUIETS>() weight 3x LTC
09-04-15 jo piece_values diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 19430 W: 3679 L: 3768 D: 11983
sprt @ 15+0.05 th 1 Values after 2nd tuning. (Fingers crossed they pass!)
08-04-15 jo spsa_piece_values diff
26330/20000 iterations
44981/40000 games played
40000 @ 30+0.05 th 1 Try a 2nd tuning session with the values from the almost passed sprt test as starting point.
08-04-15 sg update_stats diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 39971 W: 6436 L: 6345 D: 27190
sprt @ 60+0.05 th 1 LTC: simplification: update stats also in check
06-04-15 lb verif diff
LLR: -2.97 (-2.94,2.94) [-1.00,4.00]
Total: 19296 W: 3609 L: 3685 D: 12002
sprt @ 15+0.05 th 1 a bit more verification search
08-04-15 mc tuned_weights diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 17581 W: 3335 L: 3431 D: 10815
sprt @ 15+0.05 th 1 Tuned eval weights
04-04-15 mc tune_weights diff
60828/50000 iterations
112500/100000 games played
100000 @ 30+0.05 th 1 Tune eval weights
05-04-15 sg update_stats diff
LLR: 3.22 (-2.94,2.94) [-3.00,1.00]
Total: 87472 W: 16929 L: 16913 D: 53630
sprt @ 15+0.05 th 1 simplification: update stats also in check
07-04-15 Vo mPickScore2 diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 3455 W: 522 L: 608 D: 2325
sprt @ 60+0.05 th 1 I discussed with Fisherman, that it is important to have move ordering to be sync with lmr. He doesn't mind if I submit a test like his but with lmr/ordering being in sync. In fact I consider it a bug that lmr and move ordering are out of sync...like we currently have now.
05-04-15 jo piece_values diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 92431 W: 17844 L: 17672 D: 56915
sprt @ 15+0.05 th 1 Verify new piece values.
07-04-15 Ha phase diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 41822 W: 8120 L: 8128 D: 25574
sprt @ 15+0.05 th 1 Further increase game phase resolution
03-04-15 Ro ConnectedTweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 101237 W: 19567 L: 19364 D: 62306
sprt @ 15+0.05 th 1 A bonus was recently added to Connected in the 2-supported case. Use same bonus in the more general 2-connected case,
06-04-15 sn delay_split diff
ELO: -0.61 +-2.8 (95%) LOS: 33.3%
Total: 20000 W: 3278 L: 3313 D: 13409
20000 @ 15+0.05 th 4 Delay split decision at CUT nodes
06-04-15 Fi mPickScore diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 45363 W: 8759 L: 8532 D: 28072
sprt @ 15+0.05 th 1 MovePicker::score<QUIETS>() weights take 2. 3x performs even better than 2x in local testing so if it also passes STC I will go with it for LTC.
07-04-15 lb can_castle diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 12221 W: 2022 L: 1888 D: 8311
sprt @ 60+0.05 th 1 prune evasions when we can castle. forgot to change tc. sorry…
06-04-15 lb can_castle diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 20677 W: 4023 L: 3901 D: 12753
sprt @ 15+0.05 th 1 prune evasions when we can castle
05-04-15 Ha phase diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 45587 W: 8814 L: 8809 D: 27964
sprt @ 15+0.05 th 1 Increase game phase resolution
05-04-15 Ha connected diff
LLR: -4.41 (-2.94,2.94) [0.00,4.00]
Total: 34403 W: 6615 L: 6729 D: 21059
sprt @ 15+0.05 th 1 Resolve discontinuity in connected pawn values
05-04-15 jk split_depth2 diff
ELO: -1.02 +-2.0 (95%) LOS: 16.4%
Total: 40000 W: 7087 L: 7204 D: 25709
40000 @ 15+0.05 th 2 Check if going from minimumSplitDepth=4 to minimumSplitDepth=5 is a regression for 2 threads
05-04-15 Fi easyCandidate diff
LLR: -3.07 (-2.94,2.94) [0.00,4.00]
Total: 46247 W: 7378 L: 7394 D: 31475
sprt @ 60+0.05 th 1 If we use over 70% of normal time(instead of the typical 10%) to verify an easy move consider the next easy candidate generated during that time valid and don't clear it. LTC
04-04-15 lb can_castle diff
LLR: 2.95 (-2.94,2.94) [-3.50,0.50]
Total: 67877 W: 12882 L: 12904 D: 42091
sprt @ 15+0.05 th 1 prune evasions when we can castle
04-04-15 Fi mPickScore diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 51338 W: 9887 L: 9645 D: 31806
sprt @ 15+0.05 th 1 Results of tuning MovePicker::score<QUIETS>() weights.
04-04-15 Vo Salmon diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 32466 W: 5093 L: 5047 D: 22326
sprt @ 60+0.05 th 1 LTC: Increase cmh bonuses by a factor of 2. Remove history margin for cmh at lmr. Local testings seems good.
04-04-15 sn split_depth2 diff
ELO: 0.30 +-2.8 (95%) LOS: 58.2%
Total: 20000 W: 3365 L: 3348 D: 13287
20000 @ 15+0.05 th 4 Check if going from minimumSplitDepth=4 to minimumSplitDepth=5 is a regression for 4 threads
05-04-15 vi access_all_areas diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 11043 W: 2117 L: 2193 D: 6733
sprt @ 15+0.05 th 1 Take 3 - only use 'normal' mobility-counting squares.
05-04-15 vi access_all_areas diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4547 W: 836 L: 930 D: 2781
sprt @ 15+0.05 th 1 Take 2 - only give bonus if all four quarters accessible. (Still counting unsafe squares.)
05-04-15 vi access_all_areas diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 2914 W: 485 L: 582 D: 1847
sprt @ 15+0.05 th 1 Try Mindbreaker's idea of a bonus for each quarter of the board you can reach. Take 1 - include sacrifice and protection squares.
05-04-15 Fi easyCandidate diff
LLR: 2.95 (-2.94,2.94) [-1.00,4.00]
Total: 7308 W: 1463 L: 1317 D: 4528
sprt @ 15+0.05 th 1 If we use over 70% of normal time(instead of the typical 10%) to verify an easy move consider the next easy candidate generated during that time valid and don't clear it.
04-04-15 jo spsa_piece_values diff
22025/25000 iterations
44648/50000 games played
50000 @ 30+0.05 th 1 Erroneously first run was against master. Sorry. Tune with 8 moves book (book depth = 6) at intermediate tc.
04-04-15 vi hotspots diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 25914 W: 4913 L: 4950 D: 16051
sprt @ 15+0.05 th 1 Instead of naively extending the bind bonus to C6 and F6, try allowing a bind bonus (implemented as a penalty to opponent's king safety) on the hot spot near the king when in a middle game scenario (i.e. when storm and shelter are factors). Local 9" TC test produced 95% LOS, but these things are notoriously TC-specific...
04-04-15 sg bishops diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7525 W: 1394 L: 1480 D: 4651
sprt @ 15+0.05 th 1 Penalty for bishop which is strongly constrained by pawns
04-04-15 sn split_depth diff
ELO: 13.40 +-4.3 (95%) LOS: 100.0%
Total: 7525 W: 1299 L: 1009 D: 5217
20000 @ 15+0.05 th 11 Try minimumSplitDepth = 4
03-04-15 sn split_depth diff
ELO: 14.87 +-4.1 (95%) LOS: 100.0%
Total: 8509 W: 1497 L: 1133 D: 5879
20000 @ 15+0.05 th 11 Try minimumSplitDepth = 5
03-04-15 Fi mPickScoreTune diff
9647/10000 iterations
17460/20000 games played
20000 @ 30+0.05 th 1 Tune a bit more after fixing init as pointed out by Marco and it still looks to be converging.
04-04-15 Fi phaseLimits diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 11617 W: 1737 L: 1854 D: 8026
sprt @ 60+0.05 th 1 2nd and final try to widen phase limits. LTC
03-04-15 jk easybest diff
LLR: -2.96 (-2.94,2.94) [-1.00,4.00]
Total: 19895 W: 3800 L: 3874 D: 12221
sprt @ 15+0.05 th 1 Relax BestMoveChanges condition for EasyMove.
03-04-15 Vo Salmon diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 19210 W: 3734 L: 3577 D: 11899
sprt @ 15+0.05 th 1 Increase cmh bonuses by a factor of 2. Remove history margin for cmh at lmr. Local testings seems good.
03-04-15 jo rook_bonus_txt diff
ELO: -7.89 +-3.1 (95%) LOS: 0.0%
Total: 21000 W: 4213 L: 4690 D: 12097
20000 @ 9+0.05 th 1 Just a quick measure for some values found with Texel's tuning method. See also git notes.
03-04-15 Fi phaseLimits diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 3303 W: 699 L: 584 D: 2020
sprt @ 15+0.05 th 1 2nd and final try to widen phase limits.
03-04-15 Fi mPickScoreTune diff
24418/25000 iterations
49422/50000 games played
50000 @ 30+0.05 th 1 Tune history/cmh relative weights in MovePicker::score() Idea from VoyagerOne Tuna patch.
03-04-15 Ro CenterBindTweak diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 7579 W: 1381 L: 1466 D: 4732
sprt @ 15+0.05 th 1 New tweaks on blocked pawns and some centerbinds
03-04-15 jo king_psqt diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 20404 W: 3919 L: 4005 D: 12480
sprt @ 15+0.05 th 1 SPSA values for king psqt.
02-04-15 Fi phaseLimits diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 60117 W: 9509 L: 9477 D: 41131
sprt @ 60+0.05 th 1 Both of my separate phase tunings clearly widened the phase limits so test those on their own. LTC
02-04-15 sn split_depth diff
ELO: 10.43 +-2.7 (95%) LOS: 100.0%
Total: 20000 W: 3390 L: 2790 D: 13820
20000 @ 15+0.05 th 11 Try minimumSplitDepth = 6
01-04-15 la rook_psqt2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 51428 W: 9879 L: 9854 D: 31695
sprt @ 15+0.05 th 1 Last try with hand-tuned values for rook psqt
02-04-15 Vo Flounder diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 8316 W: 1530 L: 1614 D: 5172
sprt @ 15+0.05 th 1 Improve move ordering and lmr.