Stockfish Testing Queue

Finished - 24195 tests

04-10-14 sg ks_blocked diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 8408 W: 1679 L: 1763 D: 4966
sprt @ 15+0.05 th 1 Reduce king safety if king side is blocked. Same bench as master because situation is rare, but i get different bench on my 1000 positions file.
04-10-14 sn doubled_isolated diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 5583 W: 1100 L: 1192 D: 3291
sprt @ 15+0.05 th 1 Take 3 : only the case of doubled isolated pawns
03-10-14 lb opposed diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20459 W: 4181 L: 4214 D: 12064
sprt @ 15+0.05 th 1 Half endgame bonus when opposed
03-10-14 jo tune_unstop diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 36963 W: 7727 L: 7427 D: 21809
sprt @ 15+0.05 th 1 STC: Tune Unstoppable.
03-10-14 jo prefetch2 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 48250 W: 9757 L: 9686 D: 28807
sprt @ 15+0.05 th 1 Remove existing TT prefetch after committing speculative prefetch. Test for no regression.
03-10-14 sn double_isolated diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 15161 W: 2562 L: 2591 D: 10008
sprt @ 60+0.05 th 1 LTC for take 2 : full penalties for isolated pawns and backward pawns (if the pawns have any other weakness).
03-10-14 jo tune_unstop diff
ELO: 2.55 +-2.6 (95%) LOS: 97.2%
Total: 30000 W: 6777 L: 6557 D: 16666
30000 @ 7+0.05 th 1 A quick check to see, if it's worth enough to push a sprt test.
03-10-14 lb opposed diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13238 W: 2665 L: 2732 D: 7841
sprt @ 15+0.05 th 1 decrease bonus when opposed
03-10-14 My KS_defended1 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 6502 W: 1346 L: 1436 D: 3720
sprt @ 15+0.05 th 1 Include minor defended attacks in king safety evaluation (take 2)
03-10-14 My KS_defended diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1749 W: 310 L: 413 D: 1026
sprt @ 15+0.05 th 1 Include minor defended attacks in king safety evaluation
03-10-14 sn double_isolated diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 16423 W: 3424 L: 3270 D: 9729
sprt @ 15+0.05 th 1 Take 2 : full penalties for isolated pawns and backward pawns (if the pawns have any other weakness).
03-10-14 ur easy_move1 diff
LLR: -2.21 (-2.94,2.94) [-1.50,4.50]
Total: 32617 W: 6643 L: 6632 D: 19342
sprt @ 15+0.05 th 1 I removed a condition that I consider logical from last try that did not pass fast enough and I do not know if it is going to pass because maybe my intuition is wrong and I need more time for easy moves when the difference is slightly more than 0.5 pawn
03-10-14 Ro KS_Corner_2 diff
LLR: -0.22 (-2.94,2.94) [-1.50,4.50]
Total: 49090 W: 10005 L: 9876 D: 29209
sprt @ 15+0.05 th 1 Take 8: next case Kh8 and pawn g8 attacked but defended only by King
03-10-14 ur easy_move1 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 24423 W: 4894 L: 4933 D: 14596
sprt @ 15+0.05 th 1 try using previous iteration to decide if a move is an easy move including not using more than 1/3 of the time for easy moves.
03-10-14 Ro KS_Corner_2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 12553 W: 2501 L: 2573 D: 7479
sprt @ 15+0.05 th 1 Take 6: next case Kg8 and pawn h8 attacked but defended only by King
03-10-14 Ro KS_Corner_2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 3552 W: 673 L: 771 D: 2108
sprt @ 15+0.05 th 1 Take 6: next case Kg8 and pawn f8 attacked but defended only by King
03-10-14 Ro KS_Corner_2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 13840 W: 2774 L: 2843 D: 8223
sprt @ 15+0.05 th 1 Take 5: next case Kg8 and pawn h7 attacked but defended only by King
03-10-14 Ro KS_Corner_2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 11208 W: 2239 L: 2315 D: 6654
sprt @ 15+0.05 th 1 Take 4: next case Kg8 and pawn g7 attacked but defended only by King
02-10-14 sn more_connections diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2834 W: 578 L: 680 D: 1576
sprt @ 15+0.05 th 1 Count more connections between pawns : the aim is to get more continuity in the eval when a pawn duet (say, b4-c4) is pushed.
02-10-14 Ro KS_Corner_2 diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7157 W: 1518 L: 1391 D: 4248
sprt @ 15+0.05 th 1 Take 3: This time same small penalty but only when for example Kg8 and unprotected on f7
02-10-14 sn double_isolated diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 30966 W: 6258 L: 6279 D: 18429
sprt @ 15+0.05 th 1 Full penalties for isolated doubled pawns and backward doubled pawns (even if the pawns are opposed)
02-10-14 My sim_IID diff
ELO: -5.39 +-4.2 (95%) LOS: 0.6%
Total: 10000 W: 1795 L: 1950 D: 6255
10000 @ 30+0.05 th 1 Indicator of simplified internal iterative deepening depth formula
02-10-14 Ro KS_Corner_2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 31016 W: 6353 L: 6373 D: 18290
sprt @ 15+0.05 th 1 Take 2: when King in corner, say H8, and G7 is protected only by the King, add a small extra penalty.
02-10-14 lu instant_mover diff
ELO: -69.17 +-4.6 (95%) LOS: 0.0%
Total: 10000 W: 1344 L: 3309 D: 5347
10000 @ 15+0.05 th 1 Measure moving «from hash» whenever the PV move was played, unless it's twice in a row; no adjustment. Low priority and I don't mind it being canceled.
02-10-14 lb unstoppable diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 37237 W: 7440 L: 7463 D: 22334
sprt @ 15+0.05 th 1 Now we got rid of candidates, retune unstoppable (should be higher w/o candidates).
02-10-14 Ro King-safety--weighting- diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 4445 W: 882 L: 978 D: 2585
sprt @ 15+0.05 th 1 introducing a new distinction in squares around the king which are only protected by the king. Giving different weight on corners vs sides.
29-09-14 aj king_threats diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 75232 W: 13097 L: 12696 D: 49439
sprt @ 60+0.05 th 1 Use tuned values for king threats: LTC. Use SPRT[0,4] as this is more of a tuning patch.
01-10-14 Ro BishopMobility diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 41943 W: 8555 L: 8545 D: 24843
sprt @ 15+0.05 th 1 Using values identical to original weights when one Bishop, but a bit lower when two bishiops.
01-10-14 jo prefetch diff
ELO: 3.19 +-2.1 (95%) LOS: 99.9%
Total: 40000 W: 7576 L: 7209 D: 25215
40000 @ 15+0.05 th 3 Test in SMP mode as suggested by Lucas: Test Peter Österlund's prefetch patch.
01-10-14 ur simple_easy diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7264 W: 1457 L: 1544 D: 4263
sprt @ 15+0.05 th 1 More complicated try that include also 1/2 pawn difference.
01-10-14 ur simple_easy diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 8790 W: 1745 L: 1828 D: 5217
sprt @ 15+0.05 th 1 detect more easy moves by always start with 2 options in the first 1% of the search(in some cases the difference is more than 1 pawn only at depth bigger than 5)
01-10-14 Ro More-x-rays-in-King-Saf diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 9894 W: 2006 L: 2086 D: 5802
sprt @ 15+0.05 th 1 Consider for example White: Bb2 and Rh1, and Black Bg7 Kg8, we must consider Bb2 to threaten the g7 and h8 square when evaluating later the king safety. For the remaining calculations, we do not change anything to the original bitmap used for mobility calculations.
01-10-14 ur simple_easy diff
LLR: -2.18 (-2.94,2.94) [-1.50,4.50]
Total: 51432 W: 10548 L: 10483 D: 30401
sprt @ 15+0.05 th 1 same as previous patch but more aggressive reduction for easy moves(time for one pawn difference is divided by 3 and not by 2)
30-09-14 ur simple_easy diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 40756 W: 8366 L: 8145 D: 24245
sprt @ 15+0.05 th 1 Another try and this time continue the search with 2 option but use less time in case of a big difference.
30-09-14 mb pp_under_attack diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 36164 W: 7372 L: 7378 D: 21414
sprt @ 15+0.05 th 1 Reduce bonus if passed pawn is under attack and unprotected.
29-09-14 My sim_con1 diff
LLR: -0.14 (-2.94,2.94) [-3.00,1.00]
Total: 96213 W: 19461 L: 19648 D: 57104
sprt @ 15+0.05 th 1 Simplify and generalise connected pawn formula - also this way connected wing pawns are not devalued. Non-regression. (pow fix)
30-09-14 lb connected diff
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 20105 W: 3667 L: 3579 D: 12859
sprt @ 60+0.05 th 1 big simplification: merge connected and candidate
30-09-14 Ro BishopMobility diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 1757 W: 337 L: 442 D: 978
sprt @ 15+0.05 th 1 Here we got a very good bench by tweaking the "pawns on same color as Bishop" penalties.
30-09-14 pe tm_shorter_book diff
ELO: 1.13 +-2.5 (95%) LOS: 81.2%
Total: 30000 W: 6176 L: 6078 D: 17746
30000 @ 15+0.05 th 1 Now TM changes are done versus shorter book. But after change to shorter book, number of moves per game made by engine increased, and results of current tm tests may be influenced by this. So check if relevant parameter is still optimal.
30-09-14 ur simple_easy diff
LLR: 0.07 (-2.94,2.94) [-1.50,4.50]
Total: 41013 W: 8363 L: 8246 D: 24404
sprt @ 15+0.05 th 1 go back to something more similar to the first patch. The last patch fail fast and I do not know why so I did something more simple that is similar to the first patch but that has easy move only in cases that the difference is at least a pawn.
30-09-14 ur better_time1 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 3711 W: 702 L: 799 D: 2210
sprt @ 15+0.05 th 1 Third try for easy move and this time the reduction is smaller than the first formula(1/3 of the time for one pawn difference 1/2 of the time for 2 pawn difference)
30-09-14 lb connected diff
LLR: 2.96 (-2.94,2.94) [-4.00,0.00]
Total: 27118 W: 5748 L: 5689 D: 15681
sprt @ 15+0.05 th 1 big simplification: merge connected and candidate
30-09-14 ur better_time1 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7153 W: 1408 L: 1495 D: 4250
sprt @ 15+0.05 th 1 different formula
28-09-14 lb master diff
ELO: 22.80 +-1.9 (95%) LOS: 100.0%
Total: 40000 W: 7285 L: 4664 D: 28051
40000 @ 60+0.05 th 1 Regression test, standard conditions. Previous: 15.9 +/- 1.8.
30-09-14 ur better_time1 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 16274 W: 3268 L: 3330 D: 9676
sprt @ 15+0.05 th 1 fix the program not to use less time when the program changes its mind even if the difference between best move and second best move was big
29-09-14 ur better_time1 diff
LLR: -0.36 (-2.94,2.94) [-1.50,4.50]
Total: 1131 W: 208 L: 218 D: 705
sprt @ 15+0.05 th 1 use the difference between best move and second best move for better time management. Bench is the same because I do not calculate the difference during bench but only in time limit games.
29-09-14 Ro Mobility diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7218 W: 1509 L: 1597 D: 4112
sprt @ 15+0.05 th 1 Mobility v_1: mob now does not include x-rays, only real possible moves. And new mob2, weighted only 1/4 to start with, take into account potential mobility if some of our pieces were not in the way, or if pin is released.
29-09-14 jo prefetch diff
LLR: 2.96 (-2.94,2.94) [0.00,6.00]
Total: 22026 W: 3829 L: 3619 D: 14578
sprt @ 60+0.05 th 1 LTC: Test Peter Oesterlund's prefetch patch.
29-09-14 ur better_time diff
ELO: -0.10 +-3.1 (95%) LOS: 47.3%
Total: 20000 W: 4052 L: 4058 D: 11890
20000 @ 15+0.05 th 1 measure the price of calculating second best move at depth smaller than 5(If the price is not too big I want to use this information for better time management by using less time when RootMoves[0].score-RootMoves[1],score is higher at depth 4.
29-09-14 aj king_threats diff
LLR: 2.99 (-2.94,2.94) [-1.50,4.50]
Total: 20559 W: 4261 L: 4095 D: 12203
sprt @ 15+0.05 th 1 Use tuned values for king threats: STC