Skip to content

Conversation

@xoto10
Copy link
Contributor

@xoto10 xoto10 commented Jan 31, 2020

Revert 5 patches which were merged when bounds were too easy to pass:

1 Dynamic Complexity:

STC 10+0.1 https://tests.stockfishchess.org/tests/view/5e31fcacec661e2e6a340d08 :
LLR: 2.97 (-2.94,2.94) {-1.50,0.50}
Total: 38130 W: 7326 L: 7189 D: 23615
Ptnml(0-2): 677, 4346, 8843, 4545, 646

LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e32c18fec661e2e6a340d73 :
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 38675 W: 4941 L: 4866 D: 28868
Ptnml(0-2): 270, 3556, 11429, 3584, 291

3 More bonus for bestMoves on past PV nodes :

STC 10+0.1 https://tests.stockfishchess.org/tests/view/5e31fe93ec661e2e6a340d10 :
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 46100 W: 8853 L: 8727 D: 28520
Ptnml(0-2): 796, 5297, 10749, 5387, 813

LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e32c187ec661e2e6a340d71 :
LLR: 2.96 (-2.94,2.94) {-1.50,0.50}
Total: 16920 W: 2161 L: 2055 D: 12704
Ptnml(0-2): 115, 1498, 5006, 1569, 130

4 Tweak Restricted Piece Bonus :

STC 10+0.1 https://tests.stockfishchess.org/tests/view/5e31fefaec661e2e6a340d15 :
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 88328 W: 17060 L: 16997 D: 54271
Ptnml(0-2): 1536, 10446, 20169, 10422, 1581

LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e32c17aec661e2e6a340d6f :
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 34784 W: 4551 L: 4466 D: 25767
Ptnml(0-2): 255, 3279, 10061, 3345, 262

5 History update for pruned captures :

STC 10+0.1 https://tests.stockfishchess.org/tests/view/5e31ff5eec661e2e6a340d1a :
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 29541 W: 5735 L: 5588 D: 18218
Ptnml(0-2): 483, 3445, 6820, 3469, 545

LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e32c196ec661e2e6a340d75 :
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 22177 W: 2854 L: 2757 D: 16566
Ptnml(0-2): 143, 2005, 6555, 2055, 164

6 Tweak trapped rook penalty :

STC 10+0.1 https://tests.stockfishchess.org/tests/view/5e31ffb1ec661e2e6a340d1c :
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 24476 W: 4727 L: 4569 D: 15180
Ptnml(0-2): 390, 2834, 5659, 2933, 417

LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e32c19eec661e2e6a340d77 :
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 97332 W: 12492 L: 12466 D: 72374
Ptnml(0-2): 690, 9107, 28738, 9034, 720

All 5 as one simplification :
LTC 60+0.6 https://tests.stockfishchess.org/tests/view/5e334098708b13464ceea330 :
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 7829 W: 1079 L: 964 D: 5786
Ptnml(0-2): 52, 690, 2281, 781, 65

Bench: 5153165

Bench: 5153165
constexpr Score ThreatByPawnPush = S( 48, 39);
constexpr Score ThreatBySafePawn = S(173, 94);
constexpr Score TrappedRook = S( 52, 30);
constexpr Score TrappedRook = S( 52, 10);

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure if 10 is simpler than 30, but the rest of this revert does look like a simplification.

pb00068 referenced this pull request in 31m059/Stockfish Jan 31, 2020
@Vizvezdenec
Copy link
Member

well this parameter tweak was "proven" to be partially contributing to regression.
At least after this tests we get positive elo estimate, not that big but it's better than -2.5 with this patches.

bonus1 = stat_bonus(depth + 1);
bonus2 = pastPV || bestValue > beta + PawnValueMg ? bonus1 // larger bonus
bonus2 = bestValue > beta + PawnValueMg ? bonus1 // larger bonus
: stat_bonus(depth); // smaller bonus

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you fix the alignment?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure my git skills are up to doing that. This is a branch in my repo, but it's following (tracking?) Joost's branch. I can make another commit, but I don't know if I can then squash and push that since f10498b is his commit, not mine. Easier to fix it later in one of the trivial change patches I would say.

@vondele
Copy link
Member

vondele commented Jan 31, 2020

Thanks @xoto10 for creating the PR and the changelog message, and @Vizvezdenec for starting the tests. Merge in 6ccb1ca

@vondele vondele closed this Jan 31, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants