Stockfish Testing Queue

Finished - 30521 tests

15-11-01 Fis material_time diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 18526 W: 3679 L: 3469 D: 11378
sprt @ 10+0.1 th 1 Adjust time usage based on remaining material.
15-11-01 Roc RookvsMinors diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5073 W: 961 L: 1066 D: 3046
sprt @ 10+0.1 th 1 Remove empty squares controlled by opponent minors from the rook mobility area
15-10-31 mco no_timer diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 19444 W: 4061 L: 3936 D: 11447
sprt @ 2+0.05 th 7 Retire timer. Novel approach: first thread to reach limit wins (takes in account qsearch and no global counter, no now() call)
15-10-31 mco no_timer diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 15506 W: 3255 L: 3123 D: 9128
sprt @ 2+0.05 th 7 Retire timer. Full solution. Almost zero increment.
15-10-31 sg history diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 53511 W: 10354 L: 10321 D: 32836
sprt @ 10+0.1 th 1 Higher resolution for history values. See my comment on the repo. Test against passed NewStatFormula patch.
15-10-31 mbo deepest_move diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 7783 W: 1017 L: 876 D: 5890
sprt @ 60+0.4 th 7 Send information to GUI from deepest thread. LTC.
15-10-31 SC generalize_init diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6831 W: 1310 L: 1407 D: 4114
sprt @ 10+0.1 th 1 Tuned values from quadratic, game phase dependent initiative generalization. Some coefficients have gone very far away from initial values, which makes this very interesting.
15-10-31 Voy HPtweak1d diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 24016 W: 4677 L: 4696 D: 14643
sprt @ 10+0.1 th 1 Tweak 1d.
15-10-31 Voy HPtweak1c diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6401 W: 1158 L: 1256 D: 3987
sprt @ 10+0.1 th 1 Tweak 1c.
15-10-31 jos stop_iteration diff
LLR: -3.22 (-2.94,2.94) [0.00,4.00]
Total: 81370 W: 15737 L: 15619 D: 50014
sprt @ 10+0.1 th 1 Final try.
15-10-31 lbr tune diff
ELO: -47.23 +-8.9 (95%) LOS: 0.0%
Total: 2191 W: 265 L: 561 D: 1365
20000 @ 20+0.2 th 1 See what SE is worth, per se. Removed SE, and retuned LMR/LMP to (partly) compensate.
15-10-31 luc MoveEasiness diff
ELO: -3.14 +-2.4 (95%) LOS: 0.5%
Total: 30000 W: 5527 L: 5798 D: 18675
30000 @ 10+0.1 th 1 Hack upon easy move - is this worth tweaking?
15-10-31 lbr tune diff
ELO: -45.04 +-3.0 (95%) LOS: 0.0%
Total: 19540 W: 2405 L: 4924 D: 12211
20000 @ 20+0.2 th 1 Let's see how much Singular Extension is worth, without retuning LMR/LMP.
15-10-31 mco no_timer diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 7765 W: 1917 L: 1765 D: 4083
sprt @ 2+0.05 th 1 Retire timer. Novel approach: first thread to reach limit wins (takes in account qsearch and no global counter, no now() call)
15-10-31 lbr tune diff
32244/40000 iterations
65623/80000 games played
80000 @ 20+0.2 th 1 retune LMR+LMP w/o singular extension.
15-10-31 mco no_timer diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 11688 W: 2807 L: 2664 D: 6217
sprt @ 2+0.05 th 1 Retire timer. Full solution. Almost zero increment.
15-10-30 pec tm diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 47530 W: 6777 L: 6793 D: 33960
sprt @ 60+0.6 th 1 LTC. Try simple decrease of time per move. STC seems to be at least elo neutral. Problem with too much spent seems to arise at much longer tc and multiple threads. So I want to check what happens at LTC . Low throughput.
15-10-30 mbo deepest_move diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 26930 W: 4441 L: 4214 D: 18275
sprt @ 10+0.1 th 7 Send information to GUI from deepest thread.
15-10-28 Voy HPtweak diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 179867 W: 26775 L: 26245 D: 126847
sprt @ 60+0.4 th 1 Try original tweak (that failed yellow) at LTC. As h-pruning seems to scale well. Low Throughput.
15-10-30 Voy HPtweak1b diff
LLR: -2.98 (-2.94,2.94) [0.00,5.00]
Total: 29618 W: 5693 L: 5689 D: 18236
sprt @ 10+0.1 th 1 Take 2.
15-10-30 jos combo2 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 19049 W: 2704 L: 2800 D: 13545
sprt @ 60+0.4 th 1 LTC: Combo of two yellow tuning patches.
15-10-30 IIv reductions diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 3266 W: 574 L: 722 D: 1970
sprt @ 10+0.1 th 1 Today I have found a mathematical theory of LMR. I want to see how my initial guess performs on the framework before extensive local testings. In short match (200 games, 0.3 seconds per move, 3 threads) this version beat the master with 103.5 : 96.5, without any wild oscillations during the match.
15-10-30 Voy SBF3b diff
LLR: -2.97 (-2.94,2.94) [0.00,4.00]
Total: 136718 W: 26364 L: 26036 D: 84318
sprt @ 10+0.1 th 1 Use same formula as SBF3. But adjust stat weights/limits as we are dealing with bigger values. (this may help at LTC)
15-10-30 jos combo2 diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 113319 W: 21800 L: 21240 D: 70279
sprt @ 10+0.1 th 1 Combo of two yellow tuning patches.
15-10-30 SC quadratic_initiative_tu diff
46131/100000 iterations
137678/200000 games played
200000 @ 10+0.1 th 1 Fix base branch. Not my day, in any case.
15-10-30 jos easymove_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 129795 W: 24983 L: 24679 D: 80133
sprt @ 10+0.1 th 1 Take 3.
15-10-30 pec tm diff
LLR: -4.41 (-2.94,2.94) [0.00,4.00]
Total: 72115 W: 13874 L: 13853 D: 44388
sprt @ 10+0.1 th 1 Try simple decrease of time per move
15-10-30 Voy HPtweak1b diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 9313 W: 1730 L: 1816 D: 5767
sprt @ 10+0.1 th 1 Try to improve History-pruning...
15-10-30 jos stop_iteration diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 23645 W: 4495 L: 4569 D: 14581
sprt @ 10+0.1 th 1 Before eventually running a combo patch, let's first see how this is doing. Take 2, already stop at 80%.
15-10-30 Voy SBF3c diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 23058 W: 4457 L: 4533 D: 14068
sprt @ 10+0.1 th 1 Same as 3b. But remove FL bonus. Curious to see what happens.
15-10-30 jos stop_iteration diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 97063 W: 18619 L: 18315 D: 60129
sprt @ 10+0.1 th 1 Take 2. Don't start a new iteration, if we already spent more than 90% of our available time.
15-10-30 jos easymove_tweak^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 47535 W: 9074 L: 9064 D: 29397
sprt @ 10+0.1 th 1 Take 2.
15-10-30 jos easymove_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 40574 W: 7723 L: 7737 D: 25114
sprt @ 10+0.1 th 1 Allow consecutive easy moves, but spend more time.
15-10-30 lbr no_timer^ diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 2955 W: 733 L: 857 D: 1365
sprt @ 1.2+0.02 th 15 Each thread polls every 4k nodes: yes that means we poll N times more with N threads, but I bet the slowdown is not mesurable. And the more threads the more conservative we need to be, to protect ourselves against OS scheduler playing tricks on us. To test this, we need a very short tc, with many threads.
15-10-30 aji SBF3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13788 W: 1972 L: 2043 D: 9773
sprt @ 60+0.4 th 1 LTC test against master: Rescheduling VoyagerOne's test that passed STC
15-10-30 SC quadratic_initiative_tu diff
10/50000 iterations
16/100000 games played
100000 @ 10+0.1 th 1 Bugfix, thanks to Rocky640
15-10-30 Roc WeakPinned diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 27331 W: 5260 L: 5265 D: 16806
sprt @ 10+0.1 th 1 a one time penalty if we have a pinned piece which is not pawn supported
15-10-28 SC opposite_bishops_simple diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 29414 W: 5476 L: 5686 D: 18252
sprt @ 10+0.1 th 1 Variation on Stephane's idea of removing ad-hoc logic for opposite bishops. Take 2.
15-10-28 SC opposite_bishops_simple diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 25206 W: 4700 L: 4903 D: 15603
sprt @ 10+0.1 th 1 Variation on Stephane's idea of removing ad-hoc logic for opposite bishops.
15-10-30 Voy SBF3 diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 109459 W: 21288 L: 20738 D: 67433
sprt @ 10+0.1 th 1 Try: d*d+2*d-2. Test against passed SBF2.
15-10-30 Ser cutoff_time diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16881 W: 3309 L: 3360 D: 10212
sprt @ 10+0.1 th 1 Suggested by user IIvec. Same as the previous two yellow tests, but with divisor 5.
15-10-30 Ser cutoff_time_div8 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 32054 W: 6189 L: 6173 D: 19692
sprt @ 10+0.1 th 1 Same as previous yellow "cutoff_time", but with divisor 8 instead of 10. Let's see if it is better.
15-10-29 SC quadratic_initiative_tu diff
4002/50000 iterations
8403/100000 games played
100000 @ 10+0.1 th 1 Try to tune a generalization of evaluate_initiative: every contribution is now a score and also quadratic terms are considered.
15-10-27 Roc TempoVerification diff
LLR: -3.57 (-2.94,2.94) [0.00,4.00]
Total: 31234 W: 4511 L: 4603 D: 22120
sprt @ 60+0.4 th 1 Is tempo=17 better at LTC ?
15-10-29 Ser cutoff_time diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 32866 W: 6294 L: 6275 D: 20297
sprt @ 10+0.1 th 1 This is my suggestion to prevent losses on time. It uses a formula for cutoffTime so that check_time() will stop searching before it is too late even when timer thread is lagging. See also my posts in this thread: https://groups.google.com/forum/#!topic/fishcooking/kPNocZQkW-4
15-10-29 Voy SBF2 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 21225 W: 3196 L: 3000 D: 15029
sprt @ 60+0.4 th 1 LTC: Formula 2: d*d+d-1
15-10-29 Fis play_on_increment2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22337 W: 4279 L: 4306 D: 13752
sprt @ 10+0.1 th 1 Improved time management based on Leonid's patch.
15-10-29 Voy US diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19947 W: 3791 L: 3829 D: 12327
sprt @ 10+0.1 th 1 Update Stat Tweak: Note: Depth is too low for bench to change.
15-10-29 SC opposite_bishops_simple diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 7503 W: 1341 L: 1513 D: 4649
sprt @ 10+0.1 th 1 A last go with tuned values.
15-10-28 Voy Improving diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 114820 W: 22504 L: 22114 D: 70202
sprt @ 10+0.1 th 1 Increase stats bonus if improving.