| 19-06-13 |
Viz |
sdpChecks1 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 2184 W: 409 L: 563 D: 1212
|
sprt
@ 10+0.1 th 1 |
Take 1 |
| 19-06-13 |
Viz |
SpaceKD5 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 20620 W: 4499 L: 4562 D: 11559
|
sprt
@ 10+0.1 th 1 |
Take 5 |
| 19-06-12 |
pro |
ps_shelter44 |
diff
|
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 17083 W: 3746 L: 3946 D: 9391
|
sprt
@ 10+0.1 th 1 |
back to a value for evaluate_shelter. |
| 19-06-13 |
Voy |
priorBM2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 10796 W: 2336 L: 2447 D: 6013
|
sprt
@ 10+0.1 th 1 |
ver 2 |
| 19-06-13 |
Lol |
big-king-ring |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 1498 W: 297 L: 460 D: 741
|
sprt
@ 10+0.1 th 1 |
Take 1 |
| 19-06-12 |
Voy |
priorBM |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 25379 W: 5609 L: 5648 D: 14122
|
sprt
@ 10+0.1 th 1 |
stc |
| 19-06-12 |
31m |
tweak_outpost4 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 31082 W: 6912 L: 6953 D: 17217
|
sprt
@ 10+0.1 th 1 |
Small increase to the new Outpost score (note: this is larger than it seems--it is doubled for knights). |
| 19-06-12 |
pro |
ps_minorbehind3 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 18954 W: 4174 L: 4245 D: 10535
|
sprt
@ 10+0.1 th 1 |
don't reward MinorBehindPawn for rank1. |
| 19-06-12 |
pro |
ps_passed104 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 27822 W: 6182 L: 6208 D: 15432
|
sprt
@ 10+0.1 th 1 |
bonus if passed pawn advance is supported by a pawn. k+=2. |
| 19-06-12 |
Voy |
pruningT |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 16206 W: 3575 L: 3659 D: 8972
|
sprt
@ 10+0.1 th 1 |
This may fail quickly... |
| 19-06-12 |
xot |
pawnhash2 |
diff
|
LLR: -0.11 (-2.94,2.94) [0.50,4.50]
Total: 12500 W: 2816 L: 2760 D: 6924
|
sprt
@ 10+0.1 th 1 |
Priority -1 for code review. Attempt at scaling pawn hash with tt size. See comments on sg's branch https://github.com/locutus2/Stockfish/compare/a9cca5c...7d6f307
I think this uses 8x pawn hash compared to master at stc, but even more as tt scales up (so 64x at ltc). |
| 19-06-12 |
sg |
dynamic_psqt |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 8759 W: 1876 L: 1997 D: 4886
|
sprt
@ 10+0.1 th 1 |
Create dynamic piece square tables dependent on the root position. Here try to use the distance to opponent king in moves over safe squares for midgame score. |
| 19-06-12 |
pro |
ps_minorbehind3 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 8643 W: 1882 L: 2004 D: 4757
|
sprt
@ 10+0.1 th 1 |
only advanced ranks for MinorBehindPawn. rank_3. |
| 19-06-12 |
Viz |
SpaceKD3 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 7322 W: 1558 L: 1686 D: 4078
|
sprt
@ 10+0.1 th 1 |
Take 3 |
| 19-06-12 |
pro |
ps_moveimportance11 |
diff
|
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 12478 W: 2694 L: 2884 D: 6900
|
sprt
@ 10+0.1 th 1 |
v3 (replace loop with integration of move_importance). |
| 19-06-12 |
Viz |
SpaceKD2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 7956 W: 1676 L: 1801 D: 4479
|
sprt
@ 10+0.1 th 1 |
Take 2 |
| 19-06-12 |
Viz |
SpaceKD1 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 3504 W: 762 L: 912 D: 1830
|
sprt
@ 10+0.1 th 1 |
Take 1 |
| 19-06-12 |
Viz |
IndBishRest5 |
diff
|
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 20423 W: 4477 L: 4540 D: 11406
|
sprt
@ 10+0.1 th 1 |
Take 4 |
| 19-06-12 |
MJZ |
Root-Candidates2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 60148 W: 13474 L: 13339 D: 33335
|
sprt
@ 10+0.1 th 1 |
Limit LMR to 2 plies for root "best" moves. Criteria searched nodes > 4 + first move / 10 - keep skipping 4 first moves |
| 19-06-12 |
MJZ |
Root-Candidates |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 48576 W: 10877 L: 10800 D: 26899
|
sprt
@ 10+0.1 th 1 |
Limit LMR to 2 plies for root "best" moves. Criteria searched nodes > 4 + first move / 16 - keep skipping 4 first moves |
| 19-06-12 |
Viz |
IndBishRest4 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 17814 W: 3959 L: 4035 D: 9820
|
sprt
@ 10+0.1 th 1 |
Take 3 |
| 19-06-12 |
Voy |
gcPruning5b |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 33539 W: 7557 L: 7554 D: 18428
|
sprt
@ 10+0.1 th 1 |
stc |
| 19-06-12 |
Viz |
IndBishRest2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15180 W: 3328 L: 3417 D: 8435
|
sprt
@ 10+0.1 th 1 |
Take 2 |
| 19-06-12 |
sg |
Root-SkipLMR |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 6718 W: 1457 L: 1589 D: 3672
|
sprt
@ 10+0.1 th 1 |
Variant of MJZ Root-SkipLMR tests which uses logarithm of nodes because did should be depth independent. The theshold is selected to get the same skip rate as master at STC. |
| 19-06-12 |
MJZ |
Root-Candidates |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 36549 W: 8102 L: 8085 D: 20362
|
sprt
@ 10+0.1 th 1 |
Limit LMR to 2 plies for root "best" moves - to compare with skipping LMR. Criteria searched nodes > 4 + first move / 10. |
| 19-06-12 |
Voy |
gcPruning5 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 33916 W: 7527 L: 7523 D: 18866
|
sprt
@ 10+0.1 th 1 |
Another variant |
| 19-06-12 |
MJZ |
Root-SkipLMR2 |
diff
|
LLR: -0.69 (-2.94,2.94) [0.00,3.50]
Total: 10333 W: 1756 L: 1768 D: 6809
|
sprt
@ 60+0.6 th 1 |
Skip LMR for root "best" moves instead of 4 first ones. Criteria searched nodes > 4 + first move / 10, Spec LTC (I will stop it if no improvement after 10k games) |
| 19-06-12 |
Voy |
npmT |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 34374 W: 7669 L: 7663 D: 19042
|
sprt
@ 10+0.1 th 1 |
stc |
| 19-06-12 |
MJZ |
Root-SkipLMR |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 30880 W: 6858 L: 6869 D: 17153
|
sprt
@ 10+0.1 th 1 |
Skip LMR for root "best" moves instead of 4 first ones. Criteria searched nodes > 4 + first move / 9 |
| 19-06-12 |
Viz |
IndBishRest1 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 22658 W: 5093 L: 5145 D: 12420
|
sprt
@ 10+0.1 th 1 |
Take 1 |
| 19-06-12 |
Elb |
restr_weak |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 25584 W: 5675 L: 5713 D: 14196
|
sprt
@ 10+0.1 th 1 |
Restricted tweak |
| 19-06-12 |
sg |
pawnsTable |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 22767 W: 5160 L: 5040 D: 12567
|
sprt
@ 10+0.1 th 1 |
The LTC with 8 times pawn cache passed fast. For double pawn cache STC failed but i want test for this size if at STC at least no regression holds. |
| 19-06-11 |
sg |
knight_bishop_psqt |
diff
|
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 76631 W: 17126 L: 16986 D: 42519
|
sprt
@ 10+0.1 th 1 |
Combine half compensation for knight and bishop psqt. |
| 19-06-12 |
sg |
fast_sort |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15056 W: 3280 L: 3370 D: 8406
|
sprt
@ 10+0.1 th 1 |
use step = #elements / 4. |
| 19-06-12 |
Elb |
restr_weak |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 11328 W: 2468 L: 2577 D: 6283
|
sprt
@ 10+0.1 th 1 |
Take 2 |
| 19-06-12 |
sg |
fast_sort |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 10753 W: 2323 L: 2434 D: 5996
|
sprt
@ 10+0.1 th 1 |
Use for quiet move sorting one round of shellsort with step size/3 before standard insertion sort (i measured a speedup of 3.5%). |
| 19-06-11 |
sg |
knight_psqt |
diff
|
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 49741 W: 11113 L: 11080 D: 27548
|
sprt
@ 10+0.1 th 1 |
My try for psqt compensation for knights after the changed outpost calculation. |
| 19-06-12 |
MJZ |
Root-SkipLMR2 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 9180 W: 2039 L: 2159 D: 4982
|
sprt
@ 10+0.1 th 1 |
Skip LMR for root "best" moves instead of 4 first ones. Criteria searched nodes > 4 + first move / 8 |
| 19-06-07 |
sg |
pawnsTable |
diff
|
LLR: 0.31 (-2.94,2.94) [0.00,4.00]
Total: 156117 W: 27046 L: 26519 D: 102552
|
sprt
@ 60+0.6 th 1 |
Doubling size seems good for LTC but not good enough. So try a speculative LTC with even 4 times more pawn cache (this adds 5 MB memory usage per thread) (Low TP) |
| 19-06-12 |
sg |
pawnsTable |
diff
|
LLR: 2.94 (-2.94,2.94) [0.00,4.00]
Total: 26340 W: 4524 L: 4286 D: 17530
|
sprt
@ 60+0.6 th 1 |
The test with doubling sizeof pawn cache struggles (have set prio back to -1). So try a speculative LTC with even 8 times more pawn cache (this adds 12.25 MB memory usage per thread) (Low TP) |
| 19-06-12 |
pro |
ps_moveimportance11 |
diff
|
LLR: -0.97 (-2.94,2.94) [-3.00,1.00]
Total: 115453 W: 25704 L: 25989 D: 63760
|
sprt
@ 10+0.1 th 1 |
v2 of integral version. |
| 19-06-12 |
sg |
knight_psqt |
diff
|
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 80453 W: 17901 L: 17746 D: 44806
|
sprt
@ 10+0.1 th 1 |
Quarter compensation for knight psqt. |
| 19-06-12 |
Elb |
qbothflanks |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 70844 W: 15974 L: 15785 D: 39085
|
sprt
@ 10+0.1 th 1 |
Take 3 |
| 19-06-12 |
MJZ |
Root-SkipLMR2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 38929 W: 8675 L: 8646 D: 21608
|
sprt
@ 10+0.1 th 1 |
Skip LMR for root "best" moves instead of 4 first ones. Criteria searched nodes > 4 + first move / 10 |
| 19-06-12 |
Viz |
KnPSQTtw1 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 44285 W: 9914 L: 9902 D: 24469
|
sprt
@ 10+0.1 th 1 |
take 1 |
| 19-06-12 |
Viz |
BishopPawnsEg2 |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 10837 W: 2309 L: 2420 D: 6108
|
sprt
@ 10+0.1 th 1 |
Take 2 |
| 19-06-12 |
MJZ |
Root-SkipLMR |
diff
|
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 20334 W: 4460 L: 4524 D: 11350
|
sprt
@ 10+0.1 th 1 |
Skip LMR for root "best" moves instead of 4 first ones. Criteria searched nodes > 4 + first move / 12 |
| 19-06-12 |
Viz |
BishopPawnsEg1 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 15586 W: 3447 L: 3534 D: 8605
|
sprt
@ 10+0.1 th 1 |
Take 1 |
| 19-06-12 |
Fis |
QuietNoSort |
diff
|
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 30812 W: 6850 L: 6892 D: 17070
|
sprt
@ 10+0.1 th 1 |
Non functional speed optimization to save work when processing quiets. |
| 19-06-12 |
Viz |
KnPSQTtw2 |
diff
|
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 15582 W: 3446 L: 3549 D: 8587
|
sprt
@ 10+0.1 th 1 |
take 2 |