| 15-08-17 |
An |
researchStatBonus4 |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 151824 W: 25179 L: 24459 D: 102186
|
sprt @ 10+0.1 th 1 |
STC Take 4. Scale with R value from LMR instead of node depth. |
| 16-08-17 |
Vo |
lmrQSv |
diff
|
LLR: 3.56 (-2.94,2.94) [0.00,5.00]
Total: 40696 W: 7416 L: 7094 D: 26186
|
sprt @ 10+0.1 th 1 |
Different variant... (fixed patch) |
| 16-08-17 |
pb |
smp_pickbest3 |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 4878 W: 940 L: 795 D: 3143
|
sprt @ 5+0.05 th 3 |
Trying to further improve pickbest-logic by comparing nominalDepth (=effective used rootDepth for best move) instead of completedDepth. |
| 15-08-17 |
sg |
update_stats_mcp2 |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 102267 W: 18660 L: 18101 D: 65506
|
sprt @ 10+0.1 th 1 |
Only for depth < 8 |
| 15-08-17 |
lb |
stats16 |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 31489 W: 4532 L: 4295 D: 22662
|
sprt @ 40+0.4 th 1 |
16bit stats, 3rd test. Verify that we really have an elo gain with strong hash pressure. This time use longer time control, and larger hash to ensure that the hash size is big w.r.t. CPU cache sizes, and it's not an artificial effect that only works with microscopic hash sizes but doesn't scale. |
| 15-08-17 |
sg |
update_stats_mcp |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 27488 W: 5003 L: 4763 D: 17722
|
sprt @ 10+0.1 th 1 |
Even more stats bonus |
| 14-08-17 |
mc |
fix |
diff
|
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 14436 W: 2304 L: 2196 D: 9936
|
sprt @ 10+0.1 th 3 |
Fix regression introduced by "Thread code reformat". |
| 14-08-17 |
Vo |
lmrQS |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 14529 W: 2682 L: 2496 D: 9351
|
sprt @ 10+0.1 th 1 |
stc |
| 14-08-17 |
mc |
ord^ |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 17970 W: 2982 L: 2857 D: 12131
|
sprt @ 10+0.1 th 1 |
Unify captures scoring: take 2
|
| 13-08-17 |
lb |
stats16 |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 73542 W: 13058 L: 13026 D: 47458
|
sprt @ 10+0.1 th 1 |
16bit stats, 2nd test. Now verify no regression with low hash pressure. |
| 09-08-17 |
lb |
stats16 |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 258430 W: 46977 L: 45943 D: 165510
|
sprt @ 10+0.1 th 1 |
16bit stats: does reducing memory footprint by 1.2MB translate into a mesurable speed-up? let's find out in 2 tests: (1) Hash=2 (high pressure) (2) Hash=8 (low pressure). |
| 13-08-17 |
Vo |
lmrTweak3b |
diff
|
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 9630 W: 1802 L: 1636 D: 6192
|
sprt @ 10+0.1 th 1 |
stc
|
| 10-08-17 |
vd |
scoreSimp |
diff
|
LLR: 2.94 (-2.94,2.94) [-3.00,1.00]
Total: 34641 W: 4521 L: 4419 D: 25701
|
sprt @ 60+0.6 th 1 |
ltc, take 2 |
| 09-08-17 |
vd |
scoreSimp |
diff
|
LLR: 3.17 (-2.94,2.94) [-3.00,1.00]
Total: 70240 W: 12543 L: 12494 D: 45203
|
sprt @ 10+0.1 th 1 |
stc |
| 09-08-17 |
vd |
scoreSimp |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 27377 W: 4957 L: 4847 D: 17573
|
sprt @ 10+0.1 th 1 |
stc, take 2 |
| 05-08-17 |
sg |
lmr_mc_depth |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 106936 W: 19421 L: 18844 D: 68671
|
sprt @ 10+0.1 th 1 |
Combine with another of my yellow lmr tweak. Retry with now fixed formula (i inserted a term at the wrong place) |
| 07-08-17 |
Vo |
lmrTweak3 |
diff
|
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 40655 W: 7450 L: 7152 D: 26053
|
sprt @ 10+0.1 th 1 |
ver. 3 |
| 06-08-17 |
ia |
lever_count |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 20536 W: 3771 L: 3559 D: 13206
|
sprt @ 10+0.1 th 1 |
malus for having few levers |
| 05-08-17 |
II |
tmm_simple' |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 2559 W: 502 L: 368 D: 1689
|
sprt @ 80/20 th 1 |
This should be a significant improvement in x/y time controls when x is huge. Test with x=80. It's a non-functional change for x<=50. |
| 03-08-17 |
El |
reg_test |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 36987 W: 6792 L: 6510 D: 23685
|
sprt @ 10+0.1 th 1 |
Do SPRT test to test for regression. |
| 03-08-17 |
II |
tmm_simple |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 5913 W: 1217 L: 1069 D: 3627
|
sprt @ 15+0 th 1 |
For the completeness of results, let's also see sudden death performance. |
| 02-08-17 |
II |
tmm_simple |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 75356 W: 9690 L: 9640 D: 56026
|
sprt @ 60+0.6 th 1 |
LTC: Take 2: as some unstable machines are losing on time, try to increase Move Overhead. This should be negligible on higher time controls, but I'm not sure for STC. |
| 02-08-17 |
II |
tmm_simple |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 25363 W: 4658 L: 4545 D: 16160
|
sprt @ 10+0.1 th 1 |
Take 2: as some unstable machines are losing on time, try to increase Move Overhead. This should be negligible on higher time controls, but I'm not sure for STC. |
| 02-08-17 |
tt |
fix_opposite_bishops |
diff
|
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 24249 W: 4349 L: 4275 D: 15625
|
sprt @ 10+0.1 th 1 |
only affects evaluation of KBBBK, KBBBBK, ect. |
| 01-08-17 |
II |
tmm_simple |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 19377 W: 3650 L: 3526 D: 12201
|
sprt @ 40/10 th 1 |
Test also movestogo case. |
| 31-07-17 |
Vo |
yellowCombo |
diff
|
LLR: 2.94 (-2.94,2.94) [0.00,5.00]
Total: 8260 W: 1490 L: 1333 D: 5437
|
sprt @ 10+0.1 th 1 |
Combine two yellow LTC patches... |
| 30-07-17 |
Ro |
WeakUnopposed |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 32557 W: 5914 L: 5652 D: 20991
|
sprt @ 10+0.1 th 1 |
Take 3: Increase also the supported outpost |
| 27-07-17 |
Vo |
searchT |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 114130 W: 20626 L: 20085 D: 73419
|
sprt @ 10+0.1 th 1 |
stc |
| 26-07-17 |
sn |
tweak_asymmetry |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 11199 W: 2088 L: 1916 D: 7195
|
sprt @ 10+0.1 th 1 |
Tweak asymmetry measure. Tested on top of the passed "Connected3" patch. |
| 25-07-17 |
vd |
mdp |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 46576 W: 5939 L: 5851 D: 34786
|
sprt @ 60+0.6 th 1 |
ltc |
| 25-07-17 |
vd |
mdp |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 11997 W: 2172 L: 2036 D: 7789
|
sprt @ 10+0.1 th 1 |
stc |
| 24-07-17 |
sn |
Connected3 |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 48044 W: 6371 L: 6099 D: 35574
|
sprt @ 60+0.6 th 1 |
LTC: Tweak connected pawns seed[] array : +5. Tested on top of the passed patch "Connected2" by Alain. |
| 24-07-17 |
vd |
qsearchhist |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 24253 W: 4421 L: 4194 D: 15638
|
sprt @ 10+0.1 th 1 |
stc, take 3 |
| 22-07-17 |
sn |
Connected3 |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 99033 W: 17939 L: 17448 D: 63646
|
sprt @ 10+0.1 th 1 |
Tweak connected pawns seed[] array : +5. Tested on top of the passed patch "Connected2" by Alain. |
| 17-07-17 |
pb |
smp_pickbest2 |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 83333 W: 13769 L: 13320 D: 56244
|
sprt @ 5+0.05 th 5 |
Take2, taking advantage of empty framework for another try |
| 21-07-17 |
Fi |
master |
diff
|
ELO: 8.22 +-2.1 (95%) LOS: 100.0%
Total: 40000 W: 7750 L: 6804 D: 25446
|
40000 @ 10+0.1 th 1 |
Let's see what a doubling is worth with TT pressure. To be compared to
http://tests.stockfishchess.org/tests/view/5971a2f80ebc5916ff649df0
Half throughput. |
| 20-07-17 |
sg |
ttRefresh2 |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 99443 W: 12796 L: 12777 D: 73870
|
sprt @ 60+0.6 th 1 |
LTC. Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification) |
| 20-07-17 |
Ro |
Connected2 |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 24721 W: 3306 L: 3191 D: 18224
|
sprt @ 60+0.6 th 1 |
LTC |
| 21-07-17 |
Fi |
master |
diff
|
ELO: 2.30 +-2.0 (95%) LOS: 98.6%
Total: 40000 W: 7376 L: 7111 D: 25513
|
40000 @ 10+0.1 th 1 |
Measure how much elo we get from doubling TT size from 4MB to 8MB at STC. See pull #1178 |
| 21-07-17 |
Vo |
hpt3 |
diff
|
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 33795 W: 6166 L: 5898 D: 21731
|
sprt @ 10+0.1 th 1 |
stc |
| 21-07-17 |
vd |
qsearchhist |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 9840 W: 1844 L: 1677 D: 6319
|
sprt @ 10+0.1 th 1 |
STC, take 2. |
| 20-07-17 |
Ro |
Connected2 |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 19613 W: 3663 L: 3540 D: 12410
|
sprt @ 10+0.1 th 1 |
STC is struggling, try a different way |
| 19-07-17 |
sg |
ttRefresh2 |
diff
|
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 60012 W: 10753 L: 10698 D: 38561
|
sprt @ 10+0.1 th 1 |
Before going to LTC check first at STC with hash pressure (Hash=1).
Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification) |
| 19-07-17 |
sg |
ttRefresh2 |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 44271 W: 8060 L: 7979 D: 28232
|
sprt @ 10+0.1 th 1 |
Instead of full refresh increment generation. Bench is the same but differs for higher depth. (Try as simplification) |
| 18-07-17 |
Vo |
lmrT2 |
diff
|
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 15238 W: 2766 L: 2578 D: 9894
|
sprt @ 10+0.1 th 1 |
stc |
| 17-07-17 |
sg |
ttRefresh |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 71414 W: 9181 L: 9126 D: 53107
|
sprt @ 60+0.6 th 1 |
LTC: Is generation refresh during tt probing really useful? |
| 17-07-17 |
sg |
ttRefresh |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 18369 W: 3427 L: 3302 D: 11640
|
sprt @ 10+0.1 th 1 |
Is generation refresh during tt probing really useful? |
| 13-07-17 |
Fi |
PSQT |
diff
|
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 134383 W: 17486 L: 16993 D: 99904
|
sprt @ 60+0.6 th 1 |
LTC Take 2 |
| 30-06-17 |
vd |
deStackMovepick |
diff
|
LLR: 3.13 (-2.94,2.94) [-3.00,1.00]
Total: 381053 W: 68071 L: 68551 D: 244431
|
sprt @ 10+0.1 th 1 |
test for no regression |
| 12-07-17 |
Vo |
aspAlpha |
diff
|
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 87507 W: 11264 L: 11230 D: 65013
|
sprt @ 60+0.6 th 1 |
LTC: Don't modify alpha window on fail-high |