Stockfish Testing Queue

Finished - 33316 tests

15-11-30 vin tuned_levers diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19617 W: 3752 L: 3791 D: 12074
sprt @ 10+0.1 th 1 Try new values from SPSA tuning. Because there is a code change as well as new values, test at [0,5]
15-11-29 Voy CME2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 80274 W: 15535 L: 15304 D: 49435
sprt @ 10+0.1 th 1 Take 2.
15-11-30 Voy bft diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 21329 W: 4030 L: 4113 D: 13186
sprt @ 10+0.1 th 1 Small bonus formula tweak..
15-11-30 jos bad_bishop diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 61462 W: 11879 L: 11732 D: 37851
sprt @ 10+0.1 th 1 Try some handmade values. Take 2.
15-11-29 cib roadrunner diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 47040 W: 8640 L: 8563 D: 29837
sprt @ 05+0.1 th 3 Helper thread do not repeat the same depth.
15-11-29 Fis easyFix diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 10472 W: 1859 L: 1721 D: 6892
sprt @ 5+0.1 th 7 Bug fix: In case we make an easy move under SMP we can't select any other thread because ONLY the main thread has been verified to make it.
15-11-28 jos timeman_tweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 67870 W: 13047 L: 12964 D: 41859
sprt @ 10+0.1 th 1 Plan 70 moves ahead.
15-11-29 vin levers_tune diff
19369/20000 iterations
40000/40000 games played
40000 @ 20+0.2 th 1 First tuning run showed no movement for 4th rank values which we know are wrong anyway, so discard those and do 2nd run starting from the endpoint of 1st, concentrating on 5th/6th ranks.
15-11-29 jos flex_time diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 29130 W: 5552 L: 5607 D: 17971
sprt @ 10+0.1 th 1 Try to spend a little less time in the opening, so that we have more time for the midgame and the early endgame available. Take 2.
15-11-29 Voy sft diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 27716 W: 5325 L: 5385 D: 17006
sprt @ 10+0.1 th 1 Take 2.
15-11-29 jos stalemate_qsearch diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9120 W: 1675 L: 1761 D: 5684
sprt @ 10+0.1 th 1 Add stalemate detection to qsearch, but only the most practical case with the lone king. See http://talkchess.com/forum/viewtopic.php?t=58410
15-11-29 Voy lmr-cmh diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5294 W: 942 L: 1045 D: 3307
sprt @ 10+0.1 th 1 cmh bonus at lmr... if we need to research
15-11-28 Voy sft diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 54976 W: 10517 L: 10480 D: 33979
sprt @ 10+0.1 th 1 Stats formula tweak.
15-11-28 mco reset_completed diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 74769 W: 12458 L: 12733 D: 49578
sprt @ 5+0.1 th 7 Fix stale bestThread data
15-11-29 sni connected2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29923 W: 5832 L: 5824 D: 18267
sprt @ 10+0.1 th 1 Equal values in connected[] between midgame and endgame. Take 2.
15-11-29 cib avgscore diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 32895 W: 6092 L: 6075 D: 20728
sprt @ 05+0.1 th 3 The idea is to use the average score (across similar threads) to compute a better estimate for next interaction score. This patch also reduces slightly delta, from 18 to 17, in case of more than one thread.
15-11-29 aji fmc_history_cmh2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 5943 W: 1087 L: 1187 D: 3669
sprt @ 10+0.1 th 1 Don't shallow prune if either history or cmh are greater than 0: STC
15-11-29 lbr cap_reduction diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 26100 W: 4924 L: 4990 D: 16186
sprt @ 10+0.1 th 1 cap reduction: take 2/2
15-11-29 aji fmc_history diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3590 W: 625 L: 736 D: 2229
sprt @ 10+0.1 th 1 Don't shallow prune moves which have a positive history: STC
15-11-29 cib delta17 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 22273 W: 4177 L: 4257 D: 13839
sprt @ 10+0.1 th 1 Parameter tuning, for smaller aspiration window. (Single Threaded case)
15-11-29 IIv lazy_log_formula diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 23882 W: 4066 L: 4093 D: 15723
sprt @ 5+0.1 th 7 Continuous version of rohanryan's patch. Some explanations will follow on the forum.
15-11-29 lbr cap_reduction^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 19104 W: 3627 L: 3718 D: 11759
sprt @ 10+0.1 th 1 cap reduction: take 1/2
15-11-29 sni connected2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6674 W: 1231 L: 1328 D: 4115
sprt @ 10+0.1 th 1 Equal values in connected[] between midgame and endgame
15-11-29 Fis crossCompile2 diff
Pending...
1 @ 10+0.1 th 1 Just testing this compiles on fishtest.
15-11-28 sg singular diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 50744 W: 7415 L: 7354 D: 35975
sprt @ 60+0.4 th 1 LTC: Give cmh bonus for previous opponent move which causes a singular extension. Lower bonus. Take 2
15-11-29 mbo pawn_tweaks2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4260 W: 824 L: 933 D: 2503
sprt @ 10+0.1 th 1 Pawn tweaks. Doubled pawn penalties now closer to the previous values.
15-11-29 Roc PawnBreakSupport diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14898 W: 2890 L: 2950 D: 9058
sprt @ 10+0.1 th 1 Take 3
15-11-28 vin levers_tune diff
19613/20000 iterations
40000/40000 games played
40000 @ 20+0.2 th 1 Tuning Lever bonuses with separation of blocked and unblocked case. Include 4th rank since the blocked/unblocked separation further removes the cancellation effect. (re-submit with correct base branch this time, duh)
15-11-29 Roc PawnBreakSupport diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11023 W: 2115 L: 2193 D: 6715
sprt @ 10+0.1 th 1 include lever support in Mobility Area
15-11-28 adp WeightTune diff
20095/35000 iterations
40680/70000 games played
70000 @ 20+0.2 th 1 Weight Tuning, I think there would be better weights, because of recent changes in search.
15-11-28 aji fmc_history_cmh1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17954 W: 3390 L: 3437 D: 11127
sprt @ 10+0.1 th 1 Don't shallow prune moves that have a positive history and cmh: STC
15-11-28 Roc SpaceModel2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 10577 W: 2115 L: 2195 D: 6267
sprt @ 10+0.1 th 1 A different way to score the space advantage.
15-11-28 Voy stats diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11573 W: 2190 L: 2265 D: 7118
sprt @ 10+0.1 th 1 stats formula idea...colors will have equal bonuses.
15-11-27 Fis easySMP diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 35272 W: 4888 L: 4886 D: 25498
sprt @ 20+0.2 th 7 Take 2. More aggressive version reduces time to 50%. LTC
15-11-28 aji fmc_final diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 8027 W: 1296 L: 1424 D: 5307
sprt @ 25+0.25 th 1 tuned fmc values: STC(25+0.25 because local tests showed that to cover depths of 16 we need > 20+0.2)
15-11-28 sni connected2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9490 W: 1803 L: 1888 D: 5799
sprt @ 10+0.1 th 1 Raise connected[] values by 33% in midgame. Tested against the green threat weight patch.
15-11-28 mbo pawn_tweaks diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 5057 W: 987 L: 1092 D: 2978
sprt @ 10+0.1 th 1 Makes pawn PST more sensible, and changes evaluation of pawn penalties.
15-11-28 sni connected2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 2694 W: 472 L: 587 D: 1635
sprt @ 10+0.1 th 1 Raise connected[] values by 66% in midgame. Tested against the green threat weight patch.
15-11-18 luc addDepthByBF diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 46995 W: 8192 L: 8122 D: 30681
sprt @ 5+0.1 th 7 Set depth for helper threads based on the branching factor, i.e. higher when we've been iterating with less nodes. Take 2: higher depths, more similar to the current ones.
15-11-28 adp WeightTune diff
11787/35000 iterations
23957/70000 games played
70000 @ 20+0.2 th 1 Tuning of Weight parameters, since due to changes in search, I think there will be better weights for passed pawns specially.
15-11-28 vin levers_4th diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 10245 W: 1910 L: 2032 D: 6303
sprt @ 10+0.1 th 1 Try tweaked values (hence [0,4]) for lever bonus as suggested and locally tested by Vennotius
15-11-28 mbo pick_thread diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 731 W: 75 L: 234 D: 422
sprt @ 5+0.1 th 7 Don't pick bestmove from threads which fail low. I'll run this as simplification because it removes the strange "score test".
15-11-27 jos bad_bishop diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 42867 W: 6265 L: 6232 D: 30370
sprt @ 60+0.4 th 1 Bad bishop lookup table with tuned values. (Since queue is almost empty, let's test at LTC.)
15-11-27 luc addDepthByBF diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 49182 W: 8462 L: 8385 D: 32335
sprt @ 5+0.1 th 7 Set depth for helper threads based on the branching factor, i.e. higher when we've been iterating with less nodes. Final take 3, more simple and continuous.
15-11-27 Voy kpsqt diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 23369 W: 4413 L: 4488 D: 14468
sprt @ 10+0.1 th 1 KPSQT-Tweak: Fixed bench.
15-11-27 sg singular diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 19354 W: 3854 L: 3640 D: 11860
sprt @ 10+0.1 th 1 Give cmh bonus for previous opponent move which causes a singular extension. Lower bonus. Take 2
15-11-27 IIv PV_vs_nonPV diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 19123 W: 5423 L: 5513 D: 8187
sprt @ 2+0.01 th 1 If allowed, I would like to see a test on very short time controls to detect where is the main problem.
15-11-27 IIv PV_vs_nonPV diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 12680 W: 2445 L: 2559 D: 7676
sprt @ 10+0.1 th 1 There is non need for difference between PV and nonPV on depth=1. Setting so, and then increasing the difference logarithmically with depth. Testing as parameter tweak as I'm only controlling K[pv][1].
15-11-27 sg singular diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6505 W: 1185 L: 1283 D: 4037
sprt @ 10+0.1 th 1 Give cmh bonus for previous opponent move which causes a singular extension
15-11-24 aji tune_fmc_final diff
39565/40000 iterations
80000/80000 games played
80000 @ 25+0.25 th 1 tune fmc(use setRange instead of hardcoding)