Skip to content

Conversation

@snicolet
Copy link
Member

@snicolet snicolet commented Feb 26, 2021

We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at these nodes and bring better updates for the PV during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes #3369

Bench: 5037279


How to continue from there ?

  • tuning the Reductions[] array.
  • we may try to use the new DistanceFromPV metric to tweak the
    pruning/reduction/extension margins.
  • we may try to simplify away the large precondition in step 16
    since LMR is used to 98.6% of the nodes in current master.

snicolet added a commit to snicolet/Stockfish that referenced this pull request Feb 26, 2021
We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at this nodes and bring better updates for the pv during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes official-stockfish/Stockfish#3369

-------------

How to continue from there ?
 - we may try to use the new DistanceFromPV metric to tweak the
   pruning/reduction/extension margins
 - we may try to simplify away the large precondition in step 16
   since LMR is used to 98.6% of the nodes in current master.

Bench: 5037279
@snicolet snicolet force-pushed the distance_from_pv5_PR branch from 5d3afdb to 5c221dc Compare February 26, 2021 09:26
@snicolet
Copy link
Member Author

snicolet commented Feb 26, 2021

@vondele

I have some variants of this patch running at fishtest at the moment (LTC), with various
degrees of agressiveness for the extensions. If any of them pass LTC, I would like to analyse
the data and maybe update the PR.

Edit:

distance <= 2: https://tests.stockfishchess.org/tests/view/6038c1567f517a561bc4a412
distance <= 3: https://tests.stockfishchess.org/tests/view/6038bf857f517a561bc4a40f
distance <= 5: https://tests.stockfishchess.org/tests/view/6038cfd47f517a561bc4a41a
distance <= 6: https://tests.stockfishchess.org/tests/view/6038c2a57f517a561bc4a414

snicolet added a commit to snicolet/Stockfish that referenced this pull request Feb 26, 2021
We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at these nodes and bring better updates for the PV during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes official-stockfish/Stockfish#3369

-------------

How to continue from there ?
 - we may try to use the new DistanceFromPV metric to tweak the
   pruning/reduction/extension margins
 - we may try to simplify away the large precondition in step 16
   since LMR is used to 98.6% of the nodes in current master.

Bench: 5037279
@snicolet snicolet force-pushed the distance_from_pv5_PR branch from 5c221dc to 5723562 Compare February 26, 2021 12:13
We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at these nodes and bring better updates for the PV during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes official-stockfish/Stockfish#3369

Bench: 5037279

-------------

How to continue from there ?
 - tuning the Reductions[] array.
 - we may try to use the new DistanceFromPV metric to tweak the
   pruning/reduction/extension margins.
 - we may try to simplify away the large precondition in step 16
   since LMR is used to 98.6% of the nodes in current master.
@snicolet snicolet force-pushed the distance_from_pv5_PR branch from 5723562 to 5db113f Compare February 26, 2021 16:59
@snicolet
Copy link
Member Author

snicolet commented Feb 26, 2021

Current results of the LTC runs for the variants:

distance <= 2 : LLR -1.48 after 2672 games, Elo -3.9, did not start well, paused
distance <= 3 : LLR 1.19 after 58800 games, Elo +0.88, probable green or yellow
distance <= 4 : LLR 2.95 after 49608 games, Elo +1.08, passed green (this PR)
distance <= 5 : LLR 0.76 after 37560 games, Elo +0.92, probable green or yellow
distance <= 6 : LLR -1.23 after 44848 games, Elo +0.53, probable yellow

So I am fine with the version in this PR, which is in the middle of the range and has good Elo estimation.
Patch is good to merge :-)

@snicolet snicolet added the to be merged Will be merged shortly label Feb 26, 2021
@vondele
Copy link
Member

vondele commented Feb 26, 2021

looks cool, I'll merge later tonight.

@vondele vondele closed this in 0f3f5d8 Feb 26, 2021
MichaelB7 pushed a commit to MichaelB7/Stockfish that referenced this pull request Apr 13, 2021
We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at these nodes and bring better updates for the PV during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes official-stockfish#3369

Bench: 5037279
snicolet added a commit to snicolet/Stockfish that referenced this pull request May 28, 2021
We introduce a metric for each internal node in search, called DistanceFromPV.
This distance indicated how far the current node is from the principal variation.

We then use this distance to search the nodes which are close to the PV a little
deeper (up to 4 plies deeper than the PV): this improves the quality of the search
at these nodes and bring better updates for the PV during search.

STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 54936 W: 5047 L: 4850 D: 45039
Ptnml(0-2): 183, 3907, 19075, 4136, 167
https://tests.stockfishchess.org/tests/view/6037b88e7f517a561bc4a392

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 49608 W: 1880 L: 1703 D: 46025
Ptnml(0-2): 22, 1514, 21555, 1691, 22
https://tests.stockfishchess.org/tests/view/6038271b7f517a561bc4a3cb

Closes official-stockfish/Stockfish#3369

Bench: 5037279

-------------

How to continue from there ?
 - tuning the Reductions[] array.
 - we may try to use the new DistanceFromPV metric to tweak the
   pruning/reduction/extension margins.
 - we may try to simplify away the large precondition in step 16
   since LMR is used to 98.6% of the nodes in current master.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

to be merged Will be merged shortly

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants