Stockfish Testing Queue

Finished - 5228 tests

17-08-07 sg passed_pawns_span diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11729 W: 2049 L: 2125 D: 7555
sprt @ 10+0.1 th 1
Lower bonus
17-08-07 sg passed_pawns_span diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 15138 W: 2624 L: 2686 D: 9828
sprt @ 10+0.1 th 1
Try also midgame bonus for best version
17-08-07 sg passed_pawns_span diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3508 W: 582 L: 692 D: 2234
sprt @ 10+0.1 th 1
Try greater span with compensation
17-08-07 sg passed_pawns_span diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18398 W: 3314 L: 3361 D: 11723
sprt @ 10+0.1 th 1
Now with compensation
17-08-07 sg passed_pawns_span diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7759 W: 1336 L: 1428 D: 4995
sprt @ 10+0.1 th 1
Bonus for distant passed pawns
17-08-06 sg lmrPath diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18181 W: 3207 L: 3255 D: 11719
sprt @ 10+0.1 th 1
Use square of move count for path importance. Lower weight=5
17-08-05 sg lmrPath diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5925 W: 1039 L: 1139 D: 3747
sprt @ 10+0.1 th 1
Use square of move count for path importance (Park with prio -1)
17-08-05 sg lmrPath diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13570 W: 2378 L: 2446 D: 8746
sprt @ 10+0.1 th 1
Weight 400
17-08-05 sg lmrPath diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15023 W: 2673 L: 2734 D: 9616
sprt @ 10+0.1 th 1
Weight 300
17-08-04 sg lmrPath3 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 35400 W: 6270 L: 6246 D: 22884
sprt @ 10+0.1 th 1
Double weight. (Park with prio -1)
17-08-04 sg lmrPath diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 30331 W: 5388 L: 5385 D: 19558
sprt @ 10+0.1 th 1
Adjust lmr according to path importance. path importance is calculated as the negated average move count along the path to the current position expect for PV nodes where its always 0 (the maximum).
17-08-04 sg lmrPath2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13961 W: 2435 L: 2501 D: 9025
sprt @ 10+0.1 th 1
Consider for path importance only moves not on the PV.
17-08-03 sg lmr_mc diff
LLR: -3.00 (-2.94,2.94) [0.00,5.00]
Total: 84931 W: 15180 L: 14950 D: 54801
sprt @ 10+0.1 th 1
Tweak lmr formula now according to move count. Asymptotic 10% more reductions with break even move count 7. So for move count < 7 less reductions are done.
17-08-03 sg lmr_mc diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 26066 W: 4616 L: 4631 D: 16819
sprt @ 10+0.1 th 1
Sorry, forgot to push last change. The other test struggles, so try a little smaller break even. Asymptotic 10% more reductions with break even move count 6
17-08-03 sg lmr_mc diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20140 W: 3568 L: 3608 D: 12964
sprt @ 10+0.1 th 1
Try the other direction. Asymptotic 10% more reductions with break even move count 8.
17-08-03 sg lmr_mc_depth diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 40461 W: 7236 L: 7190 D: 26035
sprt @ 10+0.1 th 1
Combine with another of my yellow lmr tweak.
17-08-03 sg lmr4 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9408 W: 1613 L: 1698 D: 6097
sprt @ 10+0.1 th 1
Asymptotic 10% less reduction break even depth 10
17-08-01 sg lmr4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11176 W: 1962 L: 2040 D: 7174
sprt @ 10+0.1 th 1
Asymptotic 20% less reduction, break even depth=7.
17-07-31 sg lmr3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 32491 W: 5769 L: 5757 D: 20965
sprt @ 10+0.1 th 1
Tweak lmr formula. Asymptotic 5% less reductions with a break even point of depth=7. So for depth < 7 more reductions are done.
17-07-31 sg lmr4 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 48343 W: 8848 L: 8766 D: 30729
sprt @ 10+0.1 th 1
Asymptotic 10% less reduction, break even depth=7. (Park with prio -1)
17-07-31 sg lmr2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13562 W: 2401 L: 2469 D: 8692
sprt @ 10+0.1 th 1
Move even more break even points to leafs (ratio 3:1)
17-07-31 sg lmr diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11016 W: 1916 L: 1995 D: 7105
sprt @ 10+0.1 th 1
Half effect (weight 250).
17-07-31 sg lmr diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14187 W: 2555 L: 2620 D: 9012
sprt @ 10+0.1 th 1
Double effect (weight 1000).
17-07-30 sg lmr diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 72484 W: 13129 L: 12946 D: 46409
sprt @ 10+0.1 th 1
Move break even point even more to the leaf nodes.
17-07-30 sg lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 46260 W: 8331 L: 8260 D: 29669
sprt @ 10+0.1 th 1
Move break even point more to the leaf nodes (less reduction)
17-07-30 sg lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 37483 W: 6678 L: 6645 D: 24160
sprt @ 10+0.1 th 1
Less stats reduction near the root and more near the leafs of the search tree.
17-07-29 sg passed_pawn_distance diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19225 W: 3383 L: 3427 D: 12415
sprt @ 10+0.1 th 1
The test 'More weight for promotion square distance' struggles so try even greater weight. Test against passed Connected3''
17-07-29 sg passed_pawn_distance2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 65508 W: 11632 L: 11483 D: 42393
sprt @ 10+0.1 th 1
More weight for promotion square distance. Test against passed Connected3''
17-07-29 sg passed_pawn_distance diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 35031 W: 6214 L: 6192 D: 22625
sprt @ 10+0.1 th 1
Less weight for promotion square distance. Test against passed Connected3''
17-07-29 sg passed_pawn_distance diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 27973 W: 4982 L: 4989 D: 18002
sprt @ 10+0.1 th 1
Use promotion square in additional distance penalty. Test against passed Connected3''
17-07-28 sg passed_pawn_distance2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25105 W: 4475 L: 4494 D: 16136
sprt @ 10+0.1 th 1
Use distance from opponent king to promotion and blocking square. Test against passed Connected3''
17-07-28 sg passed_pawn_distance diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15654 W: 2754 L: 2813 D: 10087
sprt @ 10+0.1 th 1
Use distance from opponent king to promotion square instead of blocking square. Test against passed Connected3''
17-07-28 sg passed_pawn_break2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11361 W: 1956 L: 2033 D: 7372
sprt @ 10+0.1 th 1
Only one-third bonus. Test against passed Connected3''
17-07-28 sg passed_pawn_break2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 4941 W: 855 L: 959 D: 3127
sprt @ 10+0.1 th 1
Only a fifth bonus. Test against passed Connected3'' patch
17-07-27 sg passed_pawn_break diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 113695 W: 20228 L: 20577 D: 72890
sprt @ 10+0.1 th 1
Remove for passed pawn breaks the rank restriction
17-07-27 sg passed_pawn_break2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 49713 W: 8953 L: 8868 D: 31892
sprt @ 10+0.1 th 1
Allow passed pawn breaks also in our half but give only quarter bonus. Test against passed Connected3'' (+last non functional commit from master) patch from Stephane. (Park with prio -1)
17-07-20 sg ttRefresh2 diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 99443 W: 12796 L: 12777 D: 73870
sprt @ 60+0.6 th 1
LTC. Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification)
17-07-17 sg ttRefresh3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 31177 W: 5650 L: 5642 D: 19885
sprt @ 10+0.1 th 1
Do generation refresh during tt probing only if bound type is BOUND_EXACT. (Park with prio -1)
17-07-19 sg ttRefresh2 diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 60012 W: 10753 L: 10698 D: 38561
sprt @ 10+0.1 th 1
Before going to LTC check first at STC with hash pressure (Hash=1). Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification)
17-07-18 sg ttEmpty3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18434 W: 3263 L: 3310 D: 11861
sprt @ 10+0.1 th 1
Use special depth to mark empty tt entries instead of zero keys (so generation can be kept at 6 bits)
17-07-19 sg ttRefresh2 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 44271 W: 8060 L: 7979 D: 28232
sprt @ 10+0.1 th 1
Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification)
17-07-19 sg ttReplace diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 10424 W: 1817 L: 1898 D: 6709
sprt @ 10+0.1 th 1
Better score in tt replacement strategy if tt move exists.
17-07-19 sg ttReplace diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22427 W: 3996 L: 4026 D: 14405
sprt @ 10+0.1 th 1
Better score for exact bounds in tt replacement strategy. Bench are the same but differs for higher depth.
17-07-17 sg ttRefresh2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 26383 W: 4787 L: 4800 D: 16796
sprt @ 10+0.1 th 1
Do generation refresh during tt probing only if tt move exists. Bench is the same but its differs for higher depth.
17-07-19 sg ttRefresh diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 19776 W: 3495 L: 3686 D: 12595
sprt @ 10+0.1 th 1
The patch passed but check it under hash pressure (Hash=1) (See pull request #1178)
17-07-17 sg ttRefresh diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 71414 W: 9181 L: 9126 D: 53107
sprt @ 60+0.6 th 1
LTC: Is generation refresh during tt probing really useful?
17-07-18 sg ttEmpty diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4578 W: 771 L: 877 D: 2930
sprt @ 10+0.1 th 1
Use explicit empty flag for tt entries instead zero key. (shorten generation from 6 to 5 bits to have an extra bit for new flag). Fixed version.
17-07-18 sg ttRefresh_fix diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 26932 W: 4730 L: 4932 D: 17270
sprt @ 10+0.1 th 1
Retry with fix. Restore generation check in save because now its removed from probe (Thanks to mstembera)
17-07-17 sg ttRefresh2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 6026 W: 1012 L: 1112 D: 3902
sprt @ 10+0.1 th 1
Do generation refresh during tt probing only if tt depth > 0. Bench is the same but its differs for higher depth.
17-07-17 sg ttRefresh diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 18369 W: 3427 L: 3302 D: 11640
sprt @ 10+0.1 th 1
Is generation refresh during tt probing really useful?