Stockfish Testing Queue

Pending - 0 tests 0.0 hrs

None

Active - 0 tests

Finished - 70 tests

18-04-06 atu contempt_positive diff
ELO: -1.52 +-1.4 (95%) LOS: 1.5%
Total: 100000 W: 20079 L: 20516 D: 59405
100000 @ 10+0.1 th 1 Try the tuned values: the step function with {-3, 30, 53}.
18-04-05 atu tune_contempt_step diff
99012/100000 iterations
200000/200000 games played
200000 @ 10+0.1 th 1 Tune the three parameters of the step function.
18-04-05 atu contempt_positive diff
ELO: -1.40 +-1.9 (95%) LOS: 7.6%
Total: 50000 W: 9766 L: 9967 D: 30267
50000 @ 10+0.1 th 1 Use the simple step function with parameters {0, 28}.
18-04-01 atu contempt_positive diff
ELO: -0.74 +-1.9 (95%) LOS: 22.6%
Total: 50000 W: 10036 L: 10143 D: 29821
50000 @ 10+0.1 th 1 Drop the part of the dynamic contempt function with a negative argument.
18-04-01 atu contempt_positive diff
ELO: -4.40 +-1.3 (95%) LOS: 0.0%
Total: 100077 W: 18455 L: 19721 D: 61901
200000 @ 10+0.1 th 1 Drop the part of the dynamic contempt function with a positive argument. See how much it's worth.
18-03-29 atu contempt_phase diff
ELO: -1.50 +-0.9 (95%) LOS: 0.1%
Total: 200000 W: 38452 L: 39313 D: 122235
200000 @ 10+0.1 th 1 (fixed bench) 200000 more games with the updated versions of http://tests.stockfishchess.org/tests/view/5aaad7720ebc5902997fef82 to get more precise estimate.
18-03-29 atu contempt_initiative diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29143 W: 5815 L: 5809 D: 17519
sprt @ 10+0.1 th 1 Add the contempt correction after calculation of initiative.
18-03-16 atu contempt_phase diff
ELO: -3.13 +-1.9 (95%) LOS: 0.1%
Total: 50000 W: 9530 L: 9981 D: 30489
50000 @ 10+0.1 th 1 Apply the contempt correction to the whole evaluate function.
18-03-17 atu remove_dyn_contempt_val diff
ELO: -4.78 +-1.9 (95%) LOS: 0.0%
Total: 50000 W: 9100 L: 9788 D: 31112
50000 @ 10+0.1 th 1 Retest removal of dynamic contempt with specialized evaluation functions for endgames disabled.
18-03-15 atu contempt_initiative diff
ELO: -3.50 +-1.9 (95%) LOS: 0.0%
Total: 50000 W: 9379 L: 9882 D: 30739
50000 @ 10+0.1 th 1 Apply the correction to the aspiration window and to the draw score instead of to the evaluation.
18-03-15 atu contempt_initiative diff
ELO: -0.65 +-1.9 (95%) LOS: 25.4%
Total: 50000 W: 9780 L: 9873 D: 30347
50000 @ 10+0.1 th 1 Move the adding of contempt also past the calculation of phases. Trying to determine if distributing it among phases is essential for the effect.
18-03-15 atu contempt_initiative diff
ELO: 0.89 +-1.8 (95%) LOS: 83.0%
Total: 56000 W: 11331 L: 11188 D: 33481
50000 @ 10+0.1 th 1 Add the contempt correction after calculation of initiative. Trying to identify the cause of the effect of the dynamic contempt.
18-03-15 atu contempt_initiative diff
ELO: -2.50 +-1.9 (95%) LOS: 0.4%
Total: 50000 W: 9351 L: 9711 D: 30938
50000 @ 10+0.1 th 1 Use the contempt correction only in calculation of the initiative. This will test the hypothesis that the effect of the dynamic contempt is due to interaction with initiative calcuation.
18-03-13 atu remove_dyn_contempt_no_ diff
ELO: -1.90 +-1.0 (95%) LOS: 0.0%
Total: 200000 W: 40283 L: 41374 D: 118343
200000 @ 10+0.1 th 1 Remove dynamic contempt. Testing with the default value of contempt of 12 and without draw adjudication.
18-03-14 atu tune_contempt diff
47672/50000 iterations
100000/100000 games played
100000 @ 10+0.1 th 1 Tune the value of contempt in the absence of dynamic contempt. This will help selecting the base version for replacement of dynamic contempt.
18-03-12 atu aspiration diff
ELO: -1.40 +-2.2 (95%) LOS: 10.2%
Total: 40000 W: 7925 L: 8086 D: 23989
40000 @ 10+0.1 th 1 Remove dynamic contempt and use a different algorithm in calculation of the starting aspiration window.
18-03-09 atu remove_dyn_contempt_no_ diff
ELO: -3.27 +-0.9 (95%) LOS: 0.0%
Total: 200000 W: 37722 L: 39607 D: 122671
200000 @ 10+0.1 th 1 As vdbergh suggested (https://github.com/official-stockfish/Stockfish/pull/1394#issuecomment-363689037), a test is needed to make sure the dynamic contempt is not a fluke.
18-03-09 atu skip_depth diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4127 W: 831 L: 942 D: 2354
sprt @ 10+0.1 th 1 Skip every second depth in the main thread if the used time percentage <= 0.7
18-03-09 atu skip_depth diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 3436 W: 648 L: 817 D: 1971
sprt @ 10+0.1 th 1 Skip every second depth also in the main thread
18-02-17 atu tt_key64 diff
ELO: -0.94 +-2.5 (95%) LOS: 22.8%
Total: 24000 W: 3753 L: 3818 D: 16429
24000 @ 60+0.6 th 2 TT with full 64-bit keys, LTC
18-02-27 atu thread_increment diff
ELO: -0.21 +-2.5 (95%) LOS: 43.4%
Total: 30000 W: 5940 L: 5958 D: 18102
30000 @ 5+0.05 th 8 Relax the unnecessary constraint of atomicity of incrementing the nodes counter.
18-02-18 atu tt_key21 diff
ELO: -2.93 +-2.2 (95%) LOS: 0.5%
Total: 40000 W: 8565 L: 8902 D: 22533
40000 @ 10+0.1 th 1 TT with 21-bit keys
18-02-16 atu tt_key64 diff
ELO: 2.01 +-2.8 (95%) LOS: 92.0%
Total: 24000 W: 4979 L: 4840 D: 14181
24000 @ 10+0.1 th 2 TT with full 64-bit keys, cluster size 4; hash 256 MiB. Using the same versions as in the previous tests for comparable results.
18-02-15 atu tt_key64_CS2 diff
ELO: -7.20 +-2.2 (95%) LOS: 0.0%
Total: 39397 W: 7697 L: 8513 D: 23187
24000 @ 10+0.1 th 2 TT with full 64-bit keys and cluster size 2. The same test, but with 4MiB hash (the previous was 256MiB).
18-02-15 atu tt_key64 diff
ELO: -3.17 +-2.8 (95%) LOS: 1.4%
Total: 24000 W: 4841 L: 5060 D: 14099
24000 @ 10+0.1 th 2 TT with full 64-bit keys, cluster size 4; hash 4 MiB
18-02-14 atu tt_key64_CS2 diff
ELO: 0.69 +-5.3 (95%) LOS: 60.1%
Total: 6000 W: 1114 L: 1102 D: 3784
6000 @ 5+0.05 th 39 TT with full 64-bit keys and cluster size 2. The priority is 1 so that CoffeeOne's machine takes it.
18-02-14 atu tt_key64_CS2 diff
ELO: 4.43 +-2.8 (95%) LOS: 99.9%
Total: 24000 W: 5073 L: 4767 D: 14160
24000 @ 10+0.1 th 2 TT with full 64-bit keys and cluster size 2. Testing with 2 threads.
18-02-09 atu tt_key64 diff
ELO: -0.64 +-3.9 (95%) LOS: 37.3%
Total: 12000 W: 2309 L: 2331 D: 7360
12000 @ 5+0.05 th 14 Thread-safe TT with full 64-bit keys
18-02-05 atu half_time_no_adjudicati diff
ELO: -98.54 +-2.8 (95%) LOS: 0.0%
Total: 20000 W: 971 L: 6496 D: 12533
20000 @ 20+0.2 th 1 Contempt 0 TC 10+0.1 vs Contempt -20 TC 20+0.2 Testing with the same versions as in the previous tests of the series to ensure comparable results.
18-02-03 atu half_time_no_adjudicati diff
ELO: -113.24 +-3.0 (95%) LOS: 0.0%
Total: 20000 W: 1043 L: 7340 D: 11617
20000 @ 20+0.2 th 1 Contempt 0 TC 10+0.1 vs Contempt 0 TC 20+0.2
18-02-04 atu half_time_no_adjudicati diff
ELO: -129.81 +-3.2 (95%) LOS: 0.0%
Total: 20000 W: 1178 L: 8321 D: 10501
20000 @ 20+0.2 th 1 Contempt 0 TC 10+0.1 vs Contempt 20 TC 20+0.2
18-02-04 atu half_time_no_adjudicati diff
ELO: -107.00 +-2.9 (95%) LOS: 0.0%
Total: 20000 W: 958 L: 6930 D: 12112
20000 @ 20+0.2 th 1 Contempt -20 TC 10+0.1 vs Contempt 0 TC 20+0.2
18-02-03 atu half_time_no_adjudicati diff
ELO: -130.44 +-3.2 (95%) LOS: 0.0%
Total: 20000 W: 1116 L: 8291 D: 10593
20000 @ 20+0.2 th 1 Contempt 20 TC 10+0.1 vs Contempt 0 TC 20+0.2
18-02-03 atu no_adjudication diff
ELO: -0.00 +-558.1 (95%) LOS: 50.0%
Total: 3 W: 1 L: 1 D: 1
20000 @ 80+0.8 th 1 Contempt 20 vs Contempt 0 against a simulated stronger opponent.
17-05-25 atu 3fold_root diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 61259 W: 7965 L: 7897 D: 45397
sprt @ 60+0.6 th 1 (LTC) Another test for the patch which allows repeating the root position in the search tree. See #948.
17-05-22 atu 3fold_root diff
LLR: 2.94 (-2.94,2.94) [-3.00,1.00]
Total: 94622 W: 17059 L: 17064 D: 60499
sprt @ 10+0.1 th 1 Another test for the patch which allows repeating the root position in the search tree. See #948. (Sorry for the wrong bench.)
17-01-13 atu 3fold_root diff
ELO: 0.38 +-1.7 (95%) LOS: 66.8%
Total: 40000 W: 5166 L: 5122 D: 29712
40000 @ 60+0.6 th 1 Allow repeating the root position in the search tree. See #925 and #948.
17-01-02 atu tt_save diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 47800 W: 8399 L: 8636 D: 30765
sprt @ 10+0.1 th 1 Always replace an entry in the transposition entry.
16-12-30 atu tt_search diff
ELO: -32.78 +-12.4 (95%) LOS: 0.0%
Total: 1010 W: 123 L: 218 D: 669
1000 @ 10+0.1 th 3 Thread-safe TT: use values from TT only on zero positions
16-12-28 atu tt diff
ELO: -0.54 +-2.4 (95%) LOS: 33.2%
Total: 24000 W: 3613 L: 3650 D: 16737
24000 @ 10+0.1 th 3 The previous patch with 3 threads and 512MB hash.
16-12-28 atu tt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11835 W: 2068 L: 2143 D: 7624
sprt @ 10+0.1 th 1 Thread-safe TT: now remove unnecessary checks for pseudolegality and legality of TT moves.
16-12-25 atu tt diff
ELO: 1.16 +-5.8 (95%) LOS: 65.2%
Total: 3000 W: 331 L: 321 D: 2348
3000 @ 10+0.1 th 30 Testing the implementation of a thread-safe TT with 30 threads and 4GB size.
16-12-22 atu EasyMove_search diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 122841 W: 15448 L: 15457 D: 91936
sprt @ 60+0.6 th 1 (LTC) Don't clear EasyMove in ::search()
16-12-21 atu EasyMove_search diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 47719 W: 8438 L: 8362 D: 30919
sprt @ 10+0.1 th 1 Don't clear EasyMove in ::search()
16-12-20 atu tt_search diff
ELO: -16.46 +-3.7 (95%) LOS: 0.0%
Total: 9170 W: 1062 L: 1496 D: 6612
16000 @ 10+0.1 th 5 Now testing against master with 5 threads. My local tests showed +13% to speed with 2 threads.
16-12-19 atu tt_search diff
ELO: -2.00 +-2.3 (95%) LOS: 4.2%
Total: 32000 W: 5585 L: 5769 D: 20646
32000 @ 10+0.1 th 1 Thread-safe TT: at PV nodes start using values from TT only after the 3rd ply (so that it doesn't interfere with the easy move manager).
16-12-18 atu tt_search diff
ELO: -15.51 +-2.3 (95%) LOS: 0.0%
Total: 30958 W: 4919 L: 6300 D: 19739
48000 @ 10+0.1 th 1 Now that we can trust TT data, see how we can improve if we also use TT values at PV nodes
16-12-15 atu tt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11768 W: 1451 L: 1531 D: 8786
sprt @ 10+0.1 th 10 Thread-safe transposition table; ClusterSize = 4
16-12-12 atu tt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9605 W: 1199 L: 1286 D: 7120
sprt @ 10+0.1 th 10 Thread-safe transposition table; ClusterSize = 2
16-12-08 atu output_info diff
ELO: 0.16 +-2.2 (95%) LOS: 55.8%
Total: 33694 W: 5971 L: 5955 D: 21768
47000 @ 10+0.1 th 1 How much Elo will we gain if we don't show info during a search?