Stockfish Testing Queue

Finished - 29278 tests

16-07-29 Elb blocked_p2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12031 W: 2282 L: 2355 D: 7394
sprt @ 10+0.1 th 1 Take 4: use -2 factor
16-07-29 Elb blocked_p2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22851 W: 4407 L: 4432 D: 14012
sprt @ 10+0.1 th 1 Take 3: correct for pawns
16-07-29 luc knight_dist_ks diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12556 W: 2380 L: 2451 D: 7725
sprt @ 10+0.1 th 1 Larger penalty for an empty fishtest
16-07-29 Elb blocked_p2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11592 W: 2199 L: 2274 D: 7119
sprt @ 10+0.1 th 1 Take 2: use mstembera's yellow patch (http://tests.stockfishchess.org/tests/view/576f4e450ebc5972faa163c3) as base with factor 4 instead of 8
16-07-26 pec fail_high1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 36323 W: 6176 L: 6152 D: 23995
sprt @ 5+0.05 th 7 See if skipping fail highs at root works for helpers also
16-07-28 Fis twoThirdsDensity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29418 W: 4921 L: 4926 D: 19571
sprt @ 5+0.05 th 7 Use two thirds density map instead of half for skipping.
16-07-29 Voy tweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 30049 W: 5707 L: 5759 D: 18583
sprt @ 10+0.1 th 1 stc
16-07-28 jos tune_doubled diff
28928/30000 iterations
58803/60000 games played
60000 @ 20+0.2 th 1 Try to retune doubled pawn penalty per file.
16-07-28 jos doubled_file diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8946 W: 1639 L: 1726 D: 5581
sprt @ 10+0.1 th 1 Restore file-dependant penalty for doubled pawns.
16-07-28 Elb blocked_p2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 21969 W: 4149 L: 4179 D: 13641
sprt @ 10+0.1 th 1 Blocked pawns in evaluate_initiative
16-07-28 Voy cft_adj diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 16101 W: 3000 L: 3101 D: 10000
sprt @ 10+0.1 th 1 Increase cft's stat weight.
16-07-28 pec fail_high3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8761 W: 1489 L: 1577 D: 5695
sprt @ 5+0.05 th 7 Main thread skips fails highs (passed STC & LTC), helper threads resolve fail highs only if best move became different from that of main thread
16-07-28 Voy lmr-avg diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17374 W: 3186 L: 3237 D: 10951
sprt @ 10+0.1 th 1 Try averaging out the stats, excluding 0 values. Based off and against cft.
16-07-28 SC double_reductions diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 30567 W: 4182 L: 4197 D: 22188
sprt @ 60+0.6 th 1 Tuned values. LTC.
16-07-27 Fis pawnSpanInitiative2 diff
LLR: -2.93 (-2.94,2.94) [0.00,5.00]
Total: 19488 W: 3713 L: 3752 D: 12023
sprt @ 10+0.1 th 1 My final attempt at this. See commit notes.
16-07-27 Voy cft diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 32759 W: 4600 L: 4370 D: 23789
sprt @ 60+0.6 th 1 Sanity check- Since patch has been rewritten and a line of code was modify which changed the bench.
16-07-27 SC double_reductions diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 22320 W: 4378 L: 4152 D: 13790
sprt @ 10+0.1 th 1 Tuned values.
16-07-27 luc knight_dist_ks diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14274 W: 2759 L: 2822 D: 8693
sprt @ 10+0.1 th 1 Remote knight threats in king safety (tiny penalty).
16-07-27 IIv delta_new diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10473 W: 2045 L: 2125 D: 6303
sprt @ 10+0.1 th 1 Last try with formula for delta; now increasing delta with depth.
16-07-27 IIv rf_new diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8880 W: 1627 L: 1714 D: 5539
sprt @ 10+0.1 th 1 Delta/reduction combination. Also last try. Moving on after these two attempts.
16-07-27 SC double_reductions_tune diff
39869/50000 iterations
81176/100000 games played
100000 @ 10+0.1 th 1 Try to exploit double valued reductions. Reschedule as not converged.
16-07-27 pec fail_high2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 12893 W: 2139 L: 2211 D: 8543
sprt @ 5+0.05 th 7 Try to be more selective which fail highs to skip. The most critical part seems to be smp performance. I assume single thread will pass, as it benefits a lot from skipping.
16-07-27 pb0 nms_consider_hanging diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6100 W: 1103 L: 1203 D: 3794
sprt @ 10+0.1 th 1 better calculate the static eval (tempo) in the recently added condition for null move search. Don't run this directly against master as moving EvalInfo allocation to search seems to be an impact on performance, see pull 739 (fixed bench)
16-07-24 mco master diff
ELO: 49.73 +-1.7 (95%) LOS: 100.0%
Total: 40000 W: 8102 L: 2415 D: 29483
40000 @ 60+0.6 th 1 Regression test
16-07-27 luc knight_dist diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 8653 W: 1629 L: 1717 D: 5307
sprt @ 10+0.1 th 1 Hopefully fixed all bugs, with a few adjustments. (Low throughput)
16-07-27 sg move_count_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6987 W: 1247 L: 1343 D: 4397
sprt @ 10+0.1 th 1 I have collected some stats. Based on them i try it now with a depth dependent formula.
16-07-27 Voy yc2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22181 W: 4167 L: 4196 D: 13818
sprt @ 10+0.1 th 1 Yellow Combo attempt...(Fixed Bench)
16-07-27 jos fail_high diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 3081 W: 513 L: 660 D: 1908
sprt @ 10+0.1 th 1 Fully open beta already after the first fail-high. Leave alpha as is.
16-07-24 aji tune_king_safety diff
4293/30000 iterations
8867/60000 games played
60000 @ 60+0.6 th 1 Tune king safety at LTC. Run at priority -2(for an empty framework). Use much lower ck values for parameters around the lingering.
16-07-26 SC pawnSpanInitiative diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 106913 W: 20736 L: 20385 D: 65792
sprt @ 10+0.1 th 1 Very small bonus. Take 3 and call it a day.
16-07-26 pec fail_high1 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 17434 W: 3386 L: 3183 D: 10865
sprt @ 5+0.05 th 3 Helpers skip resolving fail highs . I want to see how this work separately with skip size 1. 7 thread test will put pri -1 for now
16-07-26 pec fail_high1 diff
LLR: -2.45 (-2.94,2.94) [0.00,5.00]
Total: 10000 W: 1670 L: 1732 D: 6598
sprt @ 5+0.05 th 7 Leave resolving fail high for helpers
16-07-26 SC double_reductions_tune diff
27530/30000 iterations
55730/100000 games played
100000 @ 10+0.1 th 1 Try to exploit double valued reductions.
16-07-26 pec fail_high1 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5626 W: 1025 L: 1127 D: 3474
sprt @ 5+0.05 th 3 Helpers resolve fail highs . I want to see how this work separately with skip size 1
16-07-26 lbr tuned diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 13406 W: 2503 L: 2570 D: 8333
sprt @ 10+0.1 th 1 tuned
16-07-26 Voy dangerSqs2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8485 W: 1588 L: 1677 D: 5220
sprt @ 10+0.1 th 1 One more try at this idea...
16-07-24 lbr tune diff
48816/40000 iterations
98702/100000 games played
100000 @ 20+0.2 th 1 tune relative weights of cm, fm, fm2
16-07-26 IIv rf_new diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13072 W: 2458 L: 2527 D: 8087
sprt @ 10+0.1 th 1 I plan to use this reduction formula in infinite analysis mode, because I'm stick to an idea that reductions should be lower on higher depths. Give it a try in the framework.
16-07-26 pec fail_high1 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 2031 W: 516 L: 375 D: 1140
sprt @ 5+0.05 th 1 Fast test at 5+0.05 for 1 thread to make sure that gain is still there at very short tc and to use as base for comparison with 7 thread tests
16-07-26 SC pawnSpanInitiative diff
LLR: -2.99 (-2.94,2.94) [0.00,5.00]
Total: 10836 W: 2114 L: 2194 D: 6528
sprt @ 10+0.1 th 1 Much larger bonus.
16-07-26 tvi mcpruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3022 W: 527 L: 641 D: 1854
sprt @ 10+0.1 th 1 No Movecount pruning when no ttHit
16-07-24 SC double_reductions diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 80669 W: 15286 L: 15058 D: 50325
sprt @ 10+0.1 th 1 Take 2: rounding instead of converting.
16-07-25 SC pawnSpanInitiative diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 120031 W: 23335 L: 22925 D: 73771
sprt @ 10+0.1 th 1 Pawn span initiative only for strong side.
16-07-26 luc knight_dist diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7213 W: 1343 L: 1438 D: 4432
sprt @ 10+0.1 th 1 Enemy pieces at n knight steps: I wish to test the idea before refining...
16-07-26 pec fail_high1 diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 8631 W: 1277 L: 1128 D: 6226
sprt @ 60+0.6 th 1 LTC. Skip resolving fail highs at root and partially compensate for saved time.
16-07-26 pec fail_high1 diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 6659 W: 1360 L: 1203 D: 4096
sprt @ 10+0.1 th 1 Wondering how bad it will fail if I completely remove resolving fail highs at root. Here I partially compensate for saved time.
16-07-24 SC double_reductions diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 115588 W: 22437 L: 22047 D: 71104
sprt @ 10+0.1 th 1 And take 3: adjust history contribution to have a reasonable bench. I'll wait the results of these three before deciding whether to stop or go on.
16-07-25 SC pawnSpanInitiative diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22860 W: 4378 L: 4403 D: 14079
sprt @ 10+0.1 th 1 Or, alternatively, the difference of pawn spans.
16-07-25 pec fail_high diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 11063 W: 2258 L: 2080 D: 6725
sprt @ 10+0.1 th 1 Ignore fail high at root if not failed low
16-07-25 sg move_count_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11446 W: 2206 L: 2282 D: 6958
sprt @ 10+0.1 th 1 Take 2 seems promising, so try even greater margin of 3 * pawns. Take 3