Stockfish Testing Queue

Pending - 1 tests 0.2 hrs

25-02-18 jd dynamicTempoThreat diff
Pending... (0.1 hrs)
sprt @ 10+0.1 th 1 Dynamic tempo value depending on threats, Fixed bench and rebase.

Active - 1 tests

25-02-18 jd blockedShelter diff
LLR: -1.18 (-2.94,2.94) [0.00,5.00]
Total: 7802 W: 1288 L: 1306 D: 5208
sprt @ 60+0.6 th 1 LTC for Take 2.

Finished - 119 tests

24-02-18 jd colorWeakness diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 29287 W: 6562 L: 6548 D: 16177
sprt @ 10+0.1 th 1 Small penalty if we have multiple weaknesses of the same color near our king AND we don't have the bishop of that color while our opponent does.
25-02-18 jd passedInitiative diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7061 W: 1458 L: 1554 D: 4049
sprt @ 10+0.1 th 1 Try moving passed pawn computation out of initiative, probably will fail quickly but still interesting.
25-02-18 jd blockedShelter diff
LLR: 2.94 (-2.94,2.94) [0.00,5.00]
Total: 8424 W: 1948 L: 1775 D: 4701
sprt @ 10+0.1 th 1 Take 2. Limit to h / f, a / c files and relative rank 3.
24-02-18 jd blockedShelter diff
LLR: -2.62 (-2.94,2.94) [0.00,5.00]
Total: 10224 W: 2214 L: 2280 D: 5730
sprt @ 10+0.1 th 1 Less penalty for certain blocked storm pawns, see discussion here: https://groups.google.com/forum/?fromgroups=#!searchin/fishcooking/h6$20%7Csort:date/fishcooking/p8ihDTTTDM4/S2KHhVi2CAAJ
23-02-18 jd kingShelter diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11643 W: 2597 L: 2671 D: 6375
sprt @ 10+0.1 th 1 Take 2 on shelter dependent on king rank. Double penalties / compensation.
23-02-18 jd doubled_shelter_vn diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 40281 W: 8921 L: 8853 D: 22507
sprt @ 10+0.1 th 1 A very interesting patch from nearly three years ago by Vince Negri. Passed STC and long yellow at LTC. Since king safety has been changed drastically since (e.g. it's a quadratic now, we now consider mobility danger, etc.), I think this is worth retesting.
23-02-18 jd shelterConnected diff
LLR: -1.81 (-2.94,2.94) [0.00,5.00]
Total: 5000 W: 1036 L: 1092 D: 2872
sprt @ 10+0.1 th 1 Take 2. Double penalty for lonely pawns on files near our king, change definition slightly.
23-02-18 jd kingShelter diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 25853 W: 5735 L: 5739 D: 14379
sprt @ 10+0.1 th 1 King shelter dependent on king rank. Motivated by loss to A0 here: http://www.chessgames.com/perl/chessgame?gid=1899422, from about move 17-24 SF underestimates White's position perhaps because we penalize White's king too much. Previous versions deleted before it played any games.
22-02-18 jd shelterConnected diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25567 W: 5626 L: 5631 D: 14310
sprt @ 10+0.1 th 1 ShelterWeakness currently doesn't take into account the synergy of our pawns, try to add some knowledge of this
22-02-18 jd supportedPasser3 diff
LLR: -1.65 (-2.94,2.94) [0.00,5.00]
Total: 37006 W: 8258 L: 8148 D: 20600
sprt @ 10+0.1 th 1 Take 3, final try. This time a very similar version to the one that passed STC, but a) don't require the pawn count condition and b) don't score double levers.
22-02-18 jd supportedPassers2 diff
LLR: -1.48 (-2.94,2.94) [0.00,5.00]
Total: 3837 W: 809 L: 856 D: 2172
sprt @ 10+0.1 th 1 Take 2. Fixed bench - I pasted the wrong number :-/. See discussion here: https://github.com/snicolet/Stockfish/commit/36ed7953bdc10e6ad6d6f4a230c3801236aa13e6
22-02-18 jd supportedPassers diff
LLR: -1.15 (-2.94,2.94) [0.00,5.00]
Total: 3689 W: 772 L: 805 D: 2112
sprt @ 10+0.1 th 1 Variant of Stephane's idea. See discussion here: https://github.com/snicolet/Stockfish/commit/36ed7953bdc10e6ad6d6f4a230c3801236aa13e6
21-02-18 jd passedZero diff
LLR: -1.96 (-2.94,2.94) [0.00,5.00]
Total: 5976 W: 1299 L: 1357 D: 3320
sprt @ 10+0.1 th 1 Should we allow the passed pawn evaluation for one color to be negative (due to HinderPassedPawn and/or king_proximity() being more than the bonus)?
21-02-18 jd ppBlocked diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12345 W: 2695 L: 2766 D: 6884
sprt @ 10+0.1 th 1 Give a small bonus even if our passed pawn is blocked if we attacked the blocking piece and it is not defended.
21-02-18 jd blockedBishops diff
LLR: -1.41 (-2.94,2.94) [0.00,5.00]
Total: 2216 W: 461 L: 513 D: 1242
sprt @ 10+0.1 th 1 Take 2. Simpler version, require low mobility instead of other conditions. Prio -2 at least until prev. test resolves.
21-02-18 jd blockedBishops diff
LLR: -1.62 (-2.94,2.94) [0.00,5.00]
Total: 5135 W: 1108 L: 1155 D: 2872
sprt @ 10+0.1 th 1 Attempt to solve a few small weaknesses observed in SF losses - penalty for bishop trapped behind fixed pawn centre of its own colour. Fixed bench now, sorry.
21-02-18 jd passedSecondPush diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14579 W: 3122 L: 3182 D: 8275
sprt @ 10+0.1 th 1 Another idea - consider second push of an unblocked passed pawn.
20-02-18 jd passedpawnTweak diff
LLR: -2.41 (-2.94,2.94) [0.00,5.00]
Total: 10603 W: 2269 L: 2324 D: 6010
sprt @ 10+0.1 th 1 Bench shows that on average we score passed pawns with our rook/queen behind them _less_ than passers with no rooks/queens either supporting or attacking. Try to make this more logical.
20-02-18 jd KnightEdgePasser diff
LLR: -2.56 (-2.94,2.94) [0.00,5.00]
Total: 13089 W: 2746 L: 2796 D: 7547
sprt @ 10+0.1 th 1 Only apply penalty in knight endgames, also count twice if there are passers on both A & H files.
20-02-18 jd lessDoubled diff
LLR: -1.52 (-2.94,2.94) [0.00,5.00]
Total: 8204 W: 1785 L: 1812 D: 4607
sprt @ 10+0.1 th 1 Don't score phalanx doubled, rescale penalty to compensate based on bench.
20-02-18 jd doubled diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 15821 W: 3444 L: 3498 D: 8879
sprt @ 10+0.1 th 1 Small penalty for doubled pawns regardless of whether the front pawn is supported.
19-02-18 jd quadraticKingEg diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7689 W: 1635 L: 1729 D: 4325
sprt @ 10+0.1 th 1 Try to add a small quadratic term to endgame safety.
19-02-18 jd pawnAttacks diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9040 W: 1899 L: 1986 D: 5155
sprt @ 10+0.1 th 1 Inspired by Lucas' test, try to see if distinguishing safe & unsafe pawn attacks can gain Elo.
19-02-18 jd KnightEdgePasser2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8018 W: 1693 L: 1785 D: 4540
sprt @ 10+0.1 th 1 Take 3. Score only true passers. Also, score each edge passer instead of the knights (so that if we have two edge passers the bonus is doubled).
18-02-18 jd KnightEdgePasser diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 28012 W: 6128 L: 6122 D: 15762
sprt @ 10+0.1 th 1 Half penalty. Very low priority, just parking it here while waiting for previous test to resolve.
18-02-18 jd KnightEdgePasser diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25883 W: 5704 L: 5708 D: 14471
sprt @ 10+0.1 th 1 Penalty for knights if opponent has an edge passer.
17-02-18 jd lessCheckExt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19795 W: 4294 L: 4328 D: 11173
sprt @ 10+0.1 th 1 VoyagerOne's very interesting & excellent idea with a tweak - instead of blocking all check extensions we use a condition on the see() value instead.
17-02-18 jd simplifyThreats2 diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 36770 W: 8018 L: 7926 D: 20826
sprt @ 10+0.1 th 1 Remove another unnecessary condition per Stephane's comment.
17-02-18 jd simplifyThreats diff
LLR: 0.17 (-2.94,2.94) [0.00,5.00]
Total: 98 W: 24 L: 16 D: 58
sprt @ 10+0.1 th 1 Remove defended, no functional change.
17-02-18 jd kingDangerPenalty diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6919 W: 1489 L: 1587 D: 3843
sprt @ 10+0.1 th 1 Very last try to improve on STC green / LTC yellow.
16-02-18 jd materialKingDanger diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7121 W: 1472 L: 1568 D: 4081
sprt @ 10+0.1 th 1 Compromise between STC yellow and LTC yellow - less bonus for second rook.
16-02-18 jd RookKS_Stephane diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7621 W: 1601 L: 1695 D: 4325
sprt @ 10+0.1 th 1 Multiplier 16, constant 20 per Stephane's suggestion
16-02-18 jd oppSideCastling diff
LLR: -1.52 (-2.94,2.94) [0.00,5.00]
Total: 8076 W: 1738 L: 1766 D: 4572
sprt @ 10+0.1 th 1 Adjust kingDanger by enemy king position to give a bonus if kings are on opposite sides and a penalty if they are on the same side.
16-02-18 jd tunekingDanger diff
11928/25000 iterations
25157/50000 games played
50000 @ 20+0.2 th 1 Perhaps the correct approach is to tune rather than guess :-) kingDanger seems a sensitive parameter so potentially SPSA might work well? Park with very low priority (-10) as other tests still running + framework is quite busy and evaluate whether it's worth running tomorrow.
15-02-18 jd RookKS diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 36398 W: 7985 L: 7937 D: 20476
sprt @ 10+0.1 th 1 Half bonus, as suggested by Stefan (locutus2).
15-02-18 jd materialKingDanger diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14498 W: 3125 L: 3185 D: 8188
sprt @ 10+0.1 th 1 Rebase to latest master, and experimentally consider all non-pawn pieces in kingDanger now. May fail quickly
15-02-18 jd trappedRook diff
LLR: -2.03 (-2.94,2.94) [-3.00,1.00]
Total: 12470 W: 2701 L: 2839 D: 6930
sprt @ 10+0.1 th 1 Can we simplify this any more?
15-02-18 jd RookKS2 diff
LLR: -1.82 (-2.94,2.94) [0.00,5.00]
Total: 10574 W: 2313 L: 2342 D: 5919
sprt @ 10+0.1 th 1 Previous patch isn't finding LTC easy to pass, so try a more coherent version of the idea which also considers minors which is doing very well locally.
15-02-18 jd RookKS diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 48021 W: 7994 L: 7926 D: 32101
sprt @ 60+0.6 th 1 LTC: RookKS
15-02-18 jd RookKS4 diff
LLR: -1.54 (-2.94,2.94) [0.00,5.00]
Total: 12481 W: 2733 L: 2740 D: 7008
sprt @ 10+0.1 th 1 Previous test struggling at LTC, so try compensation (-5).
15-02-18 jd RookKS2 diff
LLR: -0.43 (-2.94,2.94) [0.00,5.00]
Total: 1639 W: 363 L: 374 D: 902
sprt @ 10+0.1 th 1 Simplify previous test which passed STC.
15-02-18 jd RookKS diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 4765 W: 1120 L: 965 D: 2680
sprt @ 10+0.1 th 1 Modify kingDanger by # of rooks.
15-02-18 jd retireMinorBehindPawn2 diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 16539 W: 3571 L: 3769 D: 9199
sprt @ 10+0.1 th 1 This time compensate by increasing space weight a little.
14-02-18 jd retireFutileExtensions2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 18400 W: 4058 L: 4099 D: 10243
sprt @ 10+0.1 th 1 Take 2. Depth-dependent formula, priority -1 until previous test finishes.
14-02-18 jd reduceFutileExtensions diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 31717 W: 6954 L: 6929 D: 17834
sprt @ 10+0.1 th 1 Extra reduction when in a branch with already a lot of extensions (idea from retire_futile_extensions). 292-227-811 on quick local test at 10+0.1, might be worth a try although it's far more likely to fail quickly.
14-02-18 jd retireMinorBehindPawn diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 4744 W: 985 L: 1160 D: 2599
sprt @ 10+0.1 th 1 Try to remove MinorBehindPawn and consider such a square as a potential outpost instead. Previous torfranz test to remove it by itself failed: http://tests.stockfishchess.org/tests/view/59f6e7480ebc590ccbb89ca0
14-02-18 jd lever_danger9 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9848 W: 2117 L: 2200 D: 5531
sprt @ 10+0.1 th 1 Fixed version of idea to score only levers in our half. Very last try, previous implementation of this idea was flawed.
14-02-18 jd lever_danger8 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7973 W: 1664 L: 1756 D: 4553
sprt @ 10+0.1 th 1 Linear bonus for advanced levers + snicolet/lever_attacks4 yellow patch. Probably my last try.
14-02-18 jd lever_danger9 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 66454 W: 14727 L: 14530 D: 37197
sprt @ 10+0.1 th 1 Restricting levers to advanced pawns should allow us to add a small linear bonus as well. Prio -1 until previous test resolves.
14-02-18 jd lever_danger8 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7617 W: 1628 L: 1722 D: 4267
sprt @ 10+0.1 th 1 Framework not busy plus no other lever try succeeded yet, so try to limit levers to our side of the board.