Stockfish Testing Queue

Finished - 1016 tests

16-06-03 Elb lmr_cut2 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 25487 W: 4732 L: 4801 D: 15954
sprt @ 10+0.1 th 1 Increase reduction for cut nodes even more (test against passed LMR simplification patch)
16-06-02 Elb lmr_cut diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 15103 W: 2103 L: 1975 D: 11025
sprt @ 60+0.6 th 1 LTC: Move LMR condition to a more logical place and simplify it a bit. Run as simplification since we end up with 1 LOC less.
16-06-01 Elb lmr_cut diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 32402 W: 5967 L: 5866 D: 20569
sprt @ 10+0.1 th 1 Move LMR condition to a more logical place and simplify it a bit. Run as simplification since we end up with 1 LOC less.
16-05-31 Elb lmr_killer diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14920 W: 2673 L: 2735 D: 9512
sprt @ 10+0.1 th 1 No LMR for killers
16-05-30 Elb combo5 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 27250 W: 5046 L: 5054 D: 17150
sprt @ 10+0.1 th 1 Combo of 2 yellow patches (see commit notes)
16-05-31 Elb rook_pin diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 10957 W: 1970 L: 2049 D: 6938
sprt @ 10+0.1 th 1 Pins or discovered attacks on the opponent's rooks
16-05-27 Elb lmrt3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 83386 W: 15207 L: 14977 D: 53202
sprt @ 10+0.1 th 1 Final try on lmr improvement
16-05-27 Elb lmrt3 diff
LLR: -1.08 (-2.94,2.94) [0.00,5.00]
Total: 12191 W: 2199 L: 2193 D: 7799
sprt @ 10+0.1 th 1 Take 3
16-05-22 Elb lmrt2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 130701 W: 23994 L: 23560 D: 83147
sprt @ 10+0.1 th 1 Take 2: combine with jcalovski's idea of decreasing reduction for killers
16-05-20 Elb lmrt2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 50621 W: 9368 L: 9275 D: 31978
sprt @ 10+0.1 th 1 Don't decrease reduction for cut nodes
16-05-19 Elb combo3b diff
LLR: -1.41 (-2.94,2.94) [0.00,4.00]
Total: 65789 W: 11999 L: 11849 D: 41941
sprt @ 10+0.1 th 1 Combo patch (see commit notes). Running against passed hist_pruning patch.
16-05-20 Elb lmrt diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 26855 W: 4903 L: 4967 D: 16985
sprt @ 10+0.1 th 1 LMR parameter tweak. Test against passed hist_pruning patch.
16-05-13 Elb king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 171847 W: 31827 L: 31390 D: 108630
sprt @ 10+0.1 th 1 Update king safety values after local tuning (10K games, TC 3+0.03)
16-05-17 Elb tune_lmrpar diff
24489/20000 iterations
18329/40000 games played
40000 @ 20+0.2 th 1 Tune the parameters for calculation of rHist value.
16-05-08 Elb checks_tweak diff
LLR: -0.90 (-2.94,2.94) [-3.00,1.00]
Total: 146542 W: 26666 L: 26966 D: 92910
sprt @ 10+0.1 th 1 Use a single value for SafeCheck and OtherCheck. Run as simplification.
16-05-08 Elb qcc_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 73845 W: 13682 L: 13583 D: 46580
sprt @ 10+0.1 th 1 Local testing indicates that the QueenContactCheck value can be lowered
16-05-10 Elb qcc_tweak diff
LLR: -1.61 (-2.94,2.94) [0.00,4.00]
Total: 23881 W: 4319 L: 4323 D: 15239
sprt @ 10+0.1 th 1 Take 2 with tuned values for SafeCheck, OtherCheck and QueenContactCheck
16-05-10 Elb tune_check diff
28853/20000 iterations
40000/40000 games played
40000 @ 10+0.1 th 1 Tune the check parameters after the UnsafeChecks patch passed. Rescheduled with different start values and time control.
16-05-09 Elb tune_check diff
39670/20000 iterations
29679/40000 games played
40000 @ 20+0.2 th 1 Tune the check parameters after the UnsafeChecks patch passed. Note that I lowered the priority of http://tests.stockfishchess.org/tests/view/572f85e60ebc59301a354c6b for now.
16-05-07 Elb checks_tweak diff
LLR: -1.82 (-2.94,2.94) [0.00,4.00]
Total: 34596 W: 4608 L: 4607 D: 25381
sprt @ 60+0.6 th 1 Small increase of OtherChecked value (LTC)
16-05-08 Elb checks_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 29801 W: 5360 L: 5415 D: 19026
sprt @ 10+0.1 th 1 Take 2: try (15,0) for OtherCheck
16-05-07 Elb checks_tweak diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 22397 W: 4203 L: 3969 D: 14225
sprt @ 10+0.1 th 1 Small increase of OtherChecked value
16-05-04 Elb combo1 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 44727 W: 8038 L: 8043 D: 28646
sprt @ 10+0.1 th 1 A combo patch of two parameter tweaks (probcut and passer_tweak, see commit notes)
16-05-05 Elb combo2 diff
LLR: -2.62 (-2.94,2.94) [0.00,4.00]
Total: 15639 W: 2727 L: 2813 D: 10099
sprt @ 10+0.1 th 1 A combo patch of attack_units and Queen on 7th rank value change (see commit notes)
16-05-04 Elb lmr_var2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 5478 W: 985 L: 1087 D: 3406
sprt @ 10+0.1 th 1 An idea for the rHist formula. Test against the passed lmrUnion patch.
16-04-29 Elb lmr_denom diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 84148 W: 15359 L: 15228 D: 53561
sprt @ 10+0.1 th 1 Increase the LMR denominator value to 27K.
16-04-28 Elb lmr_denom diff
LLR: -1.30 (-2.94,2.94) [0.00,4.00]
Total: 17312 W: 2291 L: 2310 D: 12711
sprt @ 60+0.6 th 1 LMR denominator parameter tweak (23000). STC ended yellow, but I'm reasonably confident it scales better on LTC.
16-04-27 Elb lmr_denom diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 110646 W: 20114 L: 19893 D: 70639
sprt @ 10+0.1 th 1 LMR denominator parameter tweak (23000). Maybe we used the wrong start parameters for the tuning sessions.
16-04-28 Elb lmr_coeff2 diff
LLR: -2.12 (-2.94,2.94) [0.00,5.00]
Total: 11575 W: 2042 L: 2083 D: 7450
sprt @ 10+0.1 th 1 A combination of theo77186's idea of adding coefficients to the LMR formula and increasing the divider value.
16-04-27 Elb fmh_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9190 W: 1599 L: 1685 D: 5906
sprt @ 10+0.1 th 1 Dynamic LMR denominator value
16-04-26 Elb lmr_denom diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 14213 W: 2515 L: 2623 D: 9075
sprt @ 10+0.1 th 1 LMR denominator parameter tweak (20250)
16-04-26 Elb lmr_tune diff
38755/10000 iterations
19384/20000 games played
20000 @ 10+0.1 th 1 Tune LMR denominator (STC, low hash). Since the tuned value (with Hash=64) failed miserably in SPRT, do a quick tuning session with low hash (=4) to measure the difference.
16-04-25 Elb fmh_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 17971 W: 3176 L: 3271 D: 11524
sprt @ 10+0.1 th 1 LMR denominator parameter tweak (20500)
16-04-25 Elb fmh_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17491 W: 3080 L: 3131 D: 11280
sprt @ 10+0.1 th 1 Try dynamic LMR denominator selection (Take 2)
16-04-25 Elb fmh_lmr diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 12630 W: 2246 L: 2318 D: 8066
sprt @ 10+0.1 th 1 Try dynamic LMR denominator selection
16-04-24 Elb lmr_tune diff
175325/10000 iterations
20002/20000 games played
20000 @ 30+0.3 th 1 Tune LMR denominator (Medium TC)
16-04-23 Elb lmr_tune diff
83335/10000 iterations
19439/20000 games played
20000 @ 10+0.1 th 1 Tune LMR denominator (STC)
16-04-23 Elb fmh-lmrB diff
ELO: 5.49 +-5.0 (95%) LOS: 98.5%
Total: 5000 W: 713 L: 634 D: 3653
5000 @ 60+0.6 th 1 Do a quick test to check how time control dependent the LMR divider value is.
16-04-23 Elb fmh-lmrB diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 19017 W: 3363 L: 3455 D: 12199
sprt @ 10+0.1 th 1 Update divider value in LMR after tuning. Testing against passed test http://tests.stockfishchess.org/tests/view/5714d0f40ebc59301a35466b
16-04-22 Elb lmr_tune diff
126936/20000 iterations
39960/40000 games played
40000 @ 20+0.2 th 1 Now that http://tests.stockfishchess.org/tests/view/5714d0f40ebc59301a35466b has passed, try to find the optimal value for the divider.
16-04-22 Elb combo diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 31565 W: 5738 L: 5786 D: 20041
sprt @ 10+0.1 th 1 A combo patch of 2 earlier yellow patches.
16-04-22 Elb fmh2_variant diff
LLR: -1.76 (-2.94,2.94) [0.00,5.00]
Total: 6852 W: 884 L: 932 D: 5036
sprt @ 60+0.6 th 1 Test if this idea has any merit at LTC. Will stop if not positive after 10K games.
16-04-21 Elb fmh2_variant diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 36558 W: 6630 L: 6599 D: 23329
sprt @ 10+0.1 th 1 Take 2
16-04-21 Elb fmh2_variant diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 31404 W: 5725 L: 5716 D: 19963
sprt @ 10+0.1 th 1 Make LMR dependant on fmh2 and depth. Maybe this will scale better with increased time control.
16-04-20 Elb piece_checked diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 27227 W: 3629 L: 3704 D: 19894
sprt @ 60+0.6 th 1 Adjusted values for checks from Minor pieces. As this is kingsafety related and thus time control dependant, try at LTC.
16-04-19 Elb piece_checked diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 72458 W: 12980 L: 12891 D: 46587
sprt @ 10+0.1 th 1 Adjusted values for checks from Minor pieces
16-04-18 Elb scale_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 24050 W: 4353 L: 4427 D: 15270
sprt @ 10+0.1 th 1 Tweaked evaluate_scale_factor values based on local tuning. (Take 2)
16-04-18 Elb scale_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 14139 W: 2487 L: 2595 D: 9057
sprt @ 10+0.1 th 1 Tweaked evaluate_scale_factor values based on local tuning.
16-04-15 Elb attack_unit2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 84017 W: 11245 L: 11169 D: 61603
sprt @ 60+0.6 th 1 Since http://tests.stockfishchess.org/tests/view/570f879b0ebc59301a354554 failed yellow, try to see if it scales better at LTC. Will stop the test if not positive after 5000 games.
16-04-14 Elb attack_unit2 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 70471 W: 12901 L: 12816 D: 44754
sprt @ 10+0.1 th 1 Tweak attack units values (take 3)