Stockfish Testing Queue

Finished - 29420 tests

14-10-30 Fis TTtuneresult diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2740 W: 504 L: 604 D: 1632
sprt @ 15+0.05 th 1 New tuned TT replace
14-10-29 kam spsa_passed_pawns diff
19509/30000 iterations
39454/60000 games played
60000 @ 15+0.05 th 1 Tune constants in formula for passed pawns bonus. c_end = 10000, r_end = 0.0001 for both parameters. Low priority.
14-10-30 sni coordination4 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 990 W: 166 L: 272 D: 552
sprt @ 15+0.05 th 1 Rook coordination
14-10-29 Fis TTtune diff
24550/25000 iterations
50000/50000 games played
50000 @ 15+0.05 th 1 TT tuning is still converging(especially w5) so tune a bit more starting from previous best values. Pri -1
14-10-30 sni little_combinaisons9 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4263 W: 837 L: 933 D: 2493
sprt @ 15+0.05 th 1 Little combinations on under-protected pawns (penalty of half a pawn).
14-10-30 Fis fastRecapture2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 36600 W: 7400 L: 7405 D: 21795
sprt @ 15+0.05 th 1 Improved version of fast recapture. This time we also require that second best move is NOT a capture and NOT a check.
14-10-30 aji Protected_A diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 9534 W: 1876 L: 1957 D: 5701
sprt @ 15+0.05 th 1 Protected Threats tuned take 1: STC
14-10-29 gli accurate_pv diff
ELO: -1.96 +-3.1 (95%) LOS: 10.6%
Total: 18111 W: 3305 L: 3407 D: 11399
20000 @ 15+0.05 th 3 Measure accurate PV v2 with 3 threads
14-10-29 gli accurate_pv diff
ELO: -1.32 +-3.1 (95%) LOS: 20.3%
Total: 19811 W: 4041 L: 4116 D: 11654
20000 @ 15+0.05 th 1 Measure accurate PV v2 with 1 thread
14-10-28 aji pawn_threats_spsa diff
24410/25000 iterations
50000/50000 games played
50000 @ 15+0.05 th 1 Final tuning session for threats(looping across all threats instead of max threats). Tune ThreatenedByPawn and Hanging pieces. 10 parameters, 25K iterations since they are fairly elo sensitive.
14-10-28 mco no_pawn_span^ diff
LLR: 4.06 (-2.94,2.94) [-3.00,1.00]
Total: 140718 W: 28527 L: 28568 D: 83623
sprt @ 15+0.05 th 1 Retire PawnSpan: take 2
14-10-29 sni little_combinaisons9 diff
5107/5000 iterations
10000/10000 games played
10000 @ 15+0.05 th 1 Experiment with little combinations on weak pawns. Take 2: start SPSA from other initial value.
14-10-29 Roc AttackAndDefense diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1170 W: 205 L: 311 D: 654
sprt @ 15+0.05 th 1 Coordination take 3: Fixed weight for OverAttack
14-10-29 aji all_threats_tuned diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 14215 W: 2396 L: 2430 D: 9389
sprt @ 60+0.05 th 1 Add bonuses for each threat value. SPSA values for all threat bonuses: LTC
14-10-29 gli accurate_pv diff
ELO: -4.73 +-3.0 (95%) LOS: 0.1%
Total: 18823 W: 3278 L: 3534 D: 12011
20000 @ 15+0.05 th 3 Get a rough idea of ELO impact with 3 threads
14-10-29 aji all_threats_tuned diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 22454 W: 4686 L: 4515 D: 13253
sprt @ 15+0.05 th 1 Add bonuses for each threat instead of max threat value. Use SPSA tuned values for all threat bonuses : STC
14-10-29 gli accurate_pv diff
ELO: -0.63 +-3.1 (95%) LOS: 34.5%
Total: 20000 W: 4073 L: 4109 D: 11818
20000 @ 15+0.05 th 1 Get a rough idea of ELO impact in single threaded mode
14-10-29 fwi easyMove@lowCost2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7101 W: 2038 L: 2134 D: 2929
sprt @ 2+0.01 th 1 Check at short time controls check multipv at low depths and best move stability
14-10-29 Roc AttackAndDefense diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 1655 W: 299 L: 403 D: 953
sprt @ 15+0.05 th 1 coordination. Changed weights
14-10-29 mco no_single_split diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 1281 W: 174 L: 274 D: 833
sprt @ 15+0.05 th 15 Skip splitting if no slaves: must be tested with high number of threads so set at priority 1 to allocate the few capable machines, should not slow down other tests.
14-10-28 sni little_combinaisons9 diff
6527/10000 iterations
13278/20000 games played
20000 @ 15+0.05 th 1 Experiment with little combinations on weak pawns
14-10-28 Fis TTtune diff
24430/25000 iterations
50000/50000 games played
50000 @ 15+0.05 th 1 Try tuning TT replacement policy. Pri -1
14-10-29 Roc AttackAndDefense diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 1004 W: 163 L: 269 D: 572
sprt @ 15+0.05 th 1 Coordination bonus
14-10-28 pec tm_depthdep diff
ELO: 0.09 +-2.5 (95%) LOS: 52.9%
Total: 30000 W: 6000 L: 5992 D: 18008
30000 @ 15+0.05 th 1 soft stop earlier if first root move is unchanged and sufficient depth is reached
14-10-28 mco no_pawn_span diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 27293 W: 5321 L: 5531 D: 16441
sprt @ 15+0.05 th 1 Retire PawnSpan: take 3
14-10-28 Mys weak_combo diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 25363 W: 5133 L: 5170 D: 15060
sprt @ 15+0.05 th 1 Combo attacks last.
14-10-28 mco no_pawn_span diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 34605 W: 6869 L: 7094 D: 20642
sprt @ 15+0.05 th 1 Now that it seems an almost zero value of PawnSpan is the optimum, try to completely remove a good bunch of code.
14-10-27 Mys attacks diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 6283 W: 1212 L: 1302 D: 3769
sprt @ 15+0.05 th 1 Combo attacks 2.
14-10-27 jos tune_shelter diff
ELO: -3.28 +-2.6 (95%) LOS: 0.7%
Total: 30000 W: 6655 L: 6938 D: 16407
30000 @ 7+0.05 th 1 Final try on those values.
14-10-27 sg fix_see diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 13660 W: 2345 L: 2381 D: 8934
sprt @ 60+0.05 th 1 LTC: use in SEE the same piece value for knights and bishops to avoid that an exchange of a biship with a knight is considered a bad capture. This affects the move order because bad captures are sorted at the end of the move list.
14-10-27 Mys attacks diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 31327 W: 6326 L: 6346 D: 18655
sprt @ 15+0.05 th 1 Combo attacks
14-10-26 aji protected_threats_spsa_ diff
24149/25000 iterations
50000/50000 games played
50000 @ 15+0.05 th 1 Second round of spsa tuning for remaining parametes. Running at low priority. after 2K iterations at local run it seemed to have enough variation to tune on framework : STC
14-10-27 sg see_fix2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 25452 W: 5213 L: 5249 D: 14990
sprt @ 15+0.05 th 1 use in SEE the same piece value for knights and bishops to avoid that an exchange of a biship with a knight is considered a bad capture. This affects the move order because bad captures are sorted at the end of the move list. Use now standard piece value array (Take 2)
14-10-25 sg fix_see diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 149153 W: 30448 L: 29925 D: 88780
sprt @ 15+0.05 th 1 use in SEE the same piece value for knights and bishops to avoid that an exchange of a biship with a knight is considered a bad capture. This affects the move order because bad captures are sorted at the end of the move list.
14-10-26 sni pawnspan4 diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 26590 W: 4715 L: 4472 D: 17403
sprt @ 60+0.05 th 1 LTC for take 2 : medium PawnFileSpan value, independent of passed pawns
14-10-26 fwi easyMove@lowCostNew diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4525 W: 873 L: 968 D: 2684
sprt @ 15+0.05 th 1 check multipv at low depths and best move stability higher depths. Multipv part is based on one of Uri Blass's ideas. Good results locally.
14-10-26 sni pawnspan4 diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 24428 W: 5056 L: 4880 D: 14492
sprt @ 15+0.05 th 1 take 2 : medium PawnFileSpan value, independent of passed pawns
14-10-25 sni pawnspan4 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 33347 W: 6804 L: 6818 D: 19725
sprt @ 15+0.05 th 1 take 1 : half PawnFileSpan value when we have passed pawns
14-10-21 pro 3fold_1stMove diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 171356 W: 29121 L: 29556 D: 112679
sprt @ 60+0.05 th 1 LTC: fixing draw score on first repetition without affecting the search (rewritten for clarity)
14-10-25 sni pawnspan4 diff
4939/5000 iterations
9983/10000 games played
10000 @ 15+0.05 th 1 SPSA tuning for PawnFileSpan
14-10-25 luc phatm diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 12990 W: 2581 L: 2652 D: 7757
sprt @ 15+0.05 th 1 Take into account ponder hitting when determining current move's importance: take 2 (tuned with SPSA at STC)
14-10-23 luc phatm_spsa diff
24601/25000 iterations
50000/50000 games played
50000 @ 15+0.05 th 1 Tweak move importance according to whether the ponder move was previously played: initial values work great at 1000+50ms, so I schedule a low priority SPSA run at STC to hopefully better test this idea
14-10-24 jhe lmr2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 37124 W: 7582 L: 7585 D: 21957
sprt @ 15+0.05 th 1 Reduce reductions on followup moves.
14-10-24 sni threats2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 6322 W: 1227 L: 1317 D: 3778
sprt @ 15+0.05 th 1 Use max_threat() for attacks by pawns, too. Experimental run with untuned values.
14-10-24 jos tune_shelter diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 36737 W: 7522 L: 7545 D: 21670
sprt @ 15+0.05 th 1 STC: SPSA tuned values for StormDanger.
14-10-24 aji protected_threats_spsa diff
28996/30000 iterations
59446/60000 games played
60000 @ 15+0.05 th 1 Tune protected minor and major threats separately. Also loop through all threats instead of computing max threats(seems to give good results in local tests".
14-10-24 fwi timemanagement_depthbas diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18733 W: 3842 L: 3897 D: 10994
sprt @ 15+0.05 th 1 "Simplification. Apply starting at depth equals previous depth."
14-10-21 jhe sp_nodes diff
LLR: -0.09 (-2.94,2.94) [-4.00,0.00]
Total: 19000 W: 3346 L: 3415 D: 12239
sprt @ 15+0.05 th 15 15 threads.
14-10-23 jos tune_shelter diff
94084/100000 iterations
196116/200000 games played
200000 @ 10+0.05 th 1 My first SPSA session. So please double check everything. 10"+0.05 should be sufficient for tuning purpose. Since there are 12 params to tune, 200k games. Low pri.
14-10-24 fwi previousDepth_and_stabi diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 49919 W: 10177 L: 10145 D: 29597
sprt @ 15+0.05 th 1 Take 2. Apply starting at depth equals previous depth.