Lower smallnet threshold with tuned eval params #5288
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This patch uses the same material weights for the nnue amplification term and the optimism term in evaluate() STC: LLR: 2.99 (-2.94,2.94) <-1.75,0.25> Total: 83360 W: 21489 L: 21313 D: 40558 Ptnml(0-2): 303, 9934, 21056, 10058, 329 https://tests.stockfishchess.org/tests/view/664eee69928b1fb18de500d9 LTC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 192648 W: 48675 L: 48630 D: 95343 Ptnml(0-2): 82, 21484, 53161, 21501, 96 https://tests.stockfishchess.org/tests/view/664fa17aa86388d5e27d7d6e closes official-stockfish#5287 bench: 1495602
The smallnet threshold is now below the training data range of the current smallnet (simple eval diff > 1k, nn-baff1edelf90.nnue) when no pawns are on the board. Params found with spsa at 93k / 120k games at 60+06: https://tests.stockfishchess.org/tests/view/664fa166a86388d5e27d7d6b Tuned on top of: official-stockfish#5287 Passed STC: https://tests.stockfishchess.org/tests/view/664fc8b7a86388d5e27d8dac LLR: 2.96 (-2.94,2.94) <0.00,2.00> Total: 64672 W: 16731 L: 16371 D: 31570 Ptnml(0-2): 239, 7463, 16517, 7933, 184 Passed LTC: https://tests.stockfishchess.org/tests/view/664fd5f9a86388d5e27d8dfe LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 210648 W: 53489 L: 52813 D: 104346 Ptnml(0-2): 102, 23129, 58164, 23849, 80 bench 1717838
|
since this PR depends on an unmerged patch, i've applied two commits for easier rebasing/merging - the tuning patch as a separate commit on top of #5287 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The smallnet threshold is now below the training data range of
the current smallnet (simple eval diff > 1k, nn-baff1edelf90.nnue)
when no pawns are on the board.
Params found with spsa at 93k / 120k games at 60+06:
https://tests.stockfishchess.org/tests/view/664fa166a86388d5e27d7d6b
Tuned on top of: #5287
Passed STC:
https://tests.stockfishchess.org/tests/view/664fc8b7a86388d5e27d8dac
LLR: 2.96 (-2.94,2.94) <0.00,2.00>
Total: 64672 W: 16731 L: 16371 D: 31570
Ptnml(0-2): 239, 7463, 16517, 7933, 184
Passed LTC:
https://tests.stockfishchess.org/tests/view/664fd5f9a86388d5e27d8dfe
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 210648 W: 53489 L: 52813 D: 104346
Ptnml(0-2): 102, 23129, 58164, 23849, 80
bench 1717838