Stockfish Testing Queue

Finished - 4333 tests

19-07-10 sg quiet_order diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 6143 W: 1310 L: 1445 D: 3388
sprt @ 10+0.1 th 1 Scale in quiet ordering main history down with depth (break even depth=16, more slope).
19-07-10 sg quiet_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 6295 W: 1362 L: 1496 D: 3437
sprt @ 10+0.1 th 1 Scale in quiet ordering main history down with depth (break even depth=13).
19-07-10 sg capture_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 22067 W: 4861 L: 4916 D: 12290
sprt @ 10+0.1 th 1 Use piece difference / 16 for capture move ordering.
19-07-10 sg quiet_order diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 7349 W: 1599 L: 1728 D: 4022
sprt @ 10+0.1 th 1 Scale in quiet ordering main history down with depth (break even at depth=16).
19-07-10 sg goodCaptures diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 21063 W: 4663 L: 4723 D: 11677
sprt @ 10+0.1 th 1 Consider a capture with is also a discovered check always as good.
19-07-10 sg capture_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15444 W: 3376 L: 3464 D: 8604
sprt @ 10+0.1 th 1 Use a quarter of piece difference for capture move ordering.
19-07-09 sg capture_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 64781 W: 14602 L: 14443 D: 35736
sprt @ 10+0.1 th 1 Use now only an eight of captured piece - moved piece in capture move ordering.
19-07-09 sg goodCaptures diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15005 W: 3285 L: 3375 D: 8345
sprt @ 10+0.1 th 1 Consider pawn moves always as good captures.
19-07-09 sg goodCaptures diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 101991 W: 22995 L: 22650 D: 56346
sprt @ 10+0.1 th 1 Try now the other extreme and exclude all moves with negative value from good captures.
19-07-09 sg capture_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 16391 W: 3622 L: 3705 D: 9064
sprt @ 10+0.1 th 1 Use an eight of captured piece - moved piece in capture move ordering but compensate with SEE threshold.
19-07-09 sg capture_order2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 22877 W: 5052 L: 5103 D: 12722
sprt @ 10+0.1 th 1 Use an eight of difference. In addition try VoyagerOne's trick (this shows some gains but struggles) and multiply value with 8.
19-07-09 sg goodCaptures diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15788 W: 3478 L: 3564 D: 8746
sprt @ 10+0.1 th 1 Consider pawn moves (exclude non-capture promotions) always as good captures.
19-07-09 sg capture_order diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15020 W: 3378 L: 3468 D: 8174
sprt @ 10+0.1 th 1 Use also difference of captured piece - moved piece in capture move ordering.
19-07-09 sg goodCaptures diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 3411 W: 707 L: 856 D: 1848
sprt @ 10+0.1 th 1 Inspired by VoyagerOne's captureT3 test consider all moves with positive move value as good capture.
19-07-07 sg lmr_stat_score diff
LLR: -2.61 (-2.94,2.94) [0.50,4.50]
Total: 24342 W: 5501 L: 5525 D: 13316
sprt @ 10+0.1 th 1 Different stat score reductions for king/non-king moves.
19-07-07 sg lmr_stat_score diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 16288 W: 3553 L: 3637 D: 9098
sprt @ 10+0.1 th 1 Different stat score reductions for panws/non-pawns.
19-07-05 sg lmr_ttmove diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 23324 W: 5148 L: 5197 D: 12979
sprt @ 10+0.1 th 1 Less reduction if move has the same target square as the tt move.
19-07-05 sg ext_rewrite diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 36630 W: 8171 L: 8153 D: 20306
sprt @ 10+0.1 th 1 Don't consider pruned or illegal moves as extension for triggering more reduction.
19-07-05 sg ext_rewrite diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 35801 W: 8066 L: 8052 D: 19683
sprt @ 10+0.1 th 1 More reduction for a non-extension, capture move if already an extension was found.
19-07-05 sg ext_rewrite diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 33430 W: 7486 L: 7484 D: 18460
sprt @ 10+0.1 th 1 Variant of my green STC. More reduction for a non-extension, non-capture move if already a extension was found.
19-07-05 sg ext_rewrite diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 17442 W: 3856 L: 3934 D: 9652
sprt @ 10+0.1 th 1 More reduction for a non-extension move at non-PV node if already a extension was found.
19-07-05 sg ext_rewrite diff
LLR: -2.96 (-2.94,2.94) [0.00,3.50]
Total: 28091 W: 4726 L: 4823 D: 18542
sprt @ 60+0.6 th 1 LTC: Reintroduce adding extension to depth but reduce other moves if not extended by found extention.
19-07-05 sg lmr_new_squares diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 12361 W: 2680 L: 2783 D: 6898
sprt @ 10+0.1 th 1 Less reduction if current move is possible only because of previous move cleared the way.
19-07-05 sg prune_new_squares diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 8823 W: 1889 L: 2010 D: 4924
sprt @ 10+0.1 th 1 No pruning if current move is possible only because of previous move cleared the way.
19-07-05 sg ext_rewrite diff
LLR: 2.96 (-2.94,2.94) [0.50,4.50]
Total: 56541 W: 12925 L: 12476 D: 31140
sprt @ 10+0.1 th 1 Reintroduce adding extension to depth but reduce other moves if not extended by found extention.
19-07-05 sg lmr_good_squares diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 18109 W: 4069 L: 4143 D: 9897
sprt @ 10+0.1 th 1 Less reduction if piece moves to a good square. Good squares are the destination squares of tt, counter and killer moves.
19-07-05 sg lmr_ttmove diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 15501 W: 3391 L: 3479 D: 8631
sprt @ 10+0.1 th 1 More reduction if move has the same starting square as the tt move.
19-07-05 sg lmr_castle diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 18803 W: 4161 L: 4232 D: 10410
sprt @ 10+0.1 th 1 More reduction for king moves if already a castle move was searched.
19-07-05 sg lmr_castle diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 13030 W: 2838 L: 2938 D: 7254
sprt @ 10+0.1 th 1 More reduction for king moves if tt move is a castle.
19-07-05 sg ext_rewrite diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 961 W: 155 L: 318 D: 488
sprt @ 10+0.1 th 1 Experiment: instead of doing extensions increase reduction for the remaining moves if they would not be extended.
19-07-05 sg quiet_order_net_fast4 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 9122 W: 1994 L: 2114 D: 5014
sprt @ 10+0.1 th 1 Test new net (iteration 1400) against my net base version.
19-07-04 sg quiet_order_net_fast3 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 65767 W: 14698 L: 14536 D: 36533
sprt @ 10+0.1 th 1 At root with depth >= 13 but with later net (from around iteration 800). Test against my net base version.
19-07-04 sg quiet_order_net_fast2^ diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 47559 W: 10465 L: 10395 D: 26699
sprt @ 10+0.1 th 1 At root with depth >= 13 but double order value. Test against my net base version.
19-07-04 sg quiet_order_net_fast3 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 14409 W: 3131 L: 3224 D: 8054
sprt @ 10+0.1 th 1 Use net from iteration 800 with double order value. Test against my net base version.
19-07-04 sg quiet_order_net_fast2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 7636 W: 1606 L: 1733 D: 4297
sprt @ 10+0.1 th 1 At root with depth >= 13 but quadruple order value. Test against my net base version.
19-07-04 sg quiet_order_net_fast2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 19048 W: 4218 L: 4288 D: 10542
sprt @ 10+0.1 th 1 At root with depth >= 13 but half order value. Test against my net base version.
19-07-04 sg quiet_order_net_fast2^^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 17556 W: 3819 L: 3897 D: 9840
sprt @ 10+0.1 th 1 At root with lower depth >= 11. Test against my net base version.
19-07-04 sg quiet_order_net_fast2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 10219 W: 2218 L: 2332 D: 5669
sprt @ 10+0.1 th 1 Use net at all root nodes. Test against my fast base version (at root with depth >= 13). The tests so far shows that my idea is not too far from master. So first find some improvements to explore this before test again vs master (Any objections against this procedure?).
19-07-04 sg quiet_order_net_fast diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 12510 W: 2728 L: 2831 D: 6951
sprt @ 10+0.1 th 1 Thanks to @Sopel97 i have done some optimization and bugfixes which roughly doubles the speed (if we would use the net for depth >= 7). Retest with this version: only at root with depth >= 13.
19-07-04 sg quiet_order_net diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 16497 W: 3637 L: 3720 D: 9140
sprt @ 10+0.1 th 1 Use quiet order net only at the root with depth >= 13 but double quietOrderValue.
19-07-04 sg quiet_order_net diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 14759 W: 3181 L: 3273 D: 8305
sprt @ 10+0.1 th 1 Use quiet order net now for the first plies from the root with depth >= 13.
19-07-04 sg quiet_order_net2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 6620 W: 1377 L: 1509 D: 3734
sprt @ 10+0.1 th 1 Use quiet order net only at the root with depth >= 13. But use only the quietOrderValue and ignore the histories.
19-07-04 sg quiet_order_net diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 4256 W: 910 L: 1055 D: 2291
sprt @ 10+0.1 th 1 Use quiet order net now for all nodes with depth >= 13.
19-07-04 sg quiet_order_net diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 16591 W: 3693 L: 3775 D: 9123
sprt @ 10+0.1 th 1 Introduce simple neural net for quiet move ordering. Because of the heavy performance hit use the net only at the root if depth >= 13 (See more at my repo comment).
19-07-03 sg reduction_formula diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 11145 W: 2468 L: 2578 D: 6099
sprt @ 10+0.1 th 1 Increase move count by 1/24 in reduction formula (cap to MAXMOVE-1).
19-07-03 sg reduction_formula diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 38166 W: 8525 L: 8500 D: 21141
sprt @ 10+0.1 th 1 Increase move count by 1/16 in reduction formula (cap to MAXMOVE-1).
19-07-03 sg reduction_formula diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 36184 W: 8157 L: 8141 D: 19886
sprt @ 10+0.1 th 1 Increase move count by 1/32 in reduction formula (cap to MAXMOVE-1).
19-07-03 sg lmr_probcut diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 41731 W: 9380 L: 9336 D: 23015
sprt @ 10+0.1 th 1 Less reduction for quiet moves in probCut search.
19-07-03 sg reduction_formula diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 23109 W: 5204 L: 5253 D: 12652
sprt @ 10+0.1 th 1 Increase move count by 1/48 in reduction formula (cap to MAXMOVE-1).
19-07-03 sg probcut_checks diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 10377 W: 2255 L: 2368 D: 5754
sprt @ 10+0.1 th 1 Allow quiet checks above the threshold in ProbCut.