We now use per-thread dynamic contempt. This patch has the following
effects:
* for Threads=1: **non-functional**
* for Threads>1:
* with MultiPV=1: **no regression, little to no ELO gain**
* with MultiPV>1: **clear improvement over master**
First, I tried testing at standard MultiPV=1 play with [0,5] bounds.
This yielded 2 yellow and 1 red test:
5+0.05, Threads=5:
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 82689 W: 16439 L: 16190 D: 50060
http://tests.stockfishchess.org/tests/view/5aa93a5a0ebc5902952892e6
5+0.05, Threads=8:
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 27164 W: 4974 L: 4983 D: 17207
http://tests.stockfishchess.org/tests/view/5ab2639b0ebc5902a6fbefd5
5+0.5, Threads=16:
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 41396 W: 7127 L: 7082 D: 27187
http://tests.stockfishchess.org/tests/view/5ab124220ebc59029516cb62
Then, I tested with Skill Level=17 (implicitly MutliPV=4), showing
a clear improvement:
5+0.05, Threads=5:
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 3498 W: 1316 L: 1135 D: 1047
http://tests.stockfishchess.org/tests/view/5ab4b6580ebc5902932aeca2
Next, I tested the patch with MultiPV=1 again, this time checking for
non-regression ([-3, 1]):
5+0.5, Threads=5:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 65575 W: 12786 L: 12745 D: 40044
http://tests.stockfishchess.org/tests/view/5ab4e8500ebc5902932aecb3
Finally, I ran some tests with fixed number of games, checking if
reverting dynamic contempt gains more elo with Skill Level=17 (i.e.
MultiPV) than applying the "prevScore" fix and this patch. These tests
showed, that this patch gains 15 ELO when playing with Skill Level=17:
5+0.05, Threads=3, "revert dynamic contempt" vs. "WITHOUT this patch":
ELO: -11.43 +-4.1 (95%) LOS: 0.0%
Total: 20000 W: 7085 L: 7743 D: 5172
http://tests.stockfishchess.org/tests/view/5ab636450ebc590295d88536
5+0.05, Threads=3, "revert dynamic contempt" vs. "WITH this patch":
ELO: -26.42 +-4.1 (95%) LOS: 0.0%
Total: 20000 W: 6661 L: 8179 D: 5160
http://tests.stockfishchess.org/tests/view/5ab62e680ebc590295d88524
---
***FAQ***
**Why should this be commited?**
I believe that the gain for multi-thread MultiPV search is a sufficient
justification for this otherwise neutral change. I also believe this
implementation of dynamic contempt is more logical, although this may
be just my opinion.
**Why is per-thread contempt better at MultiPV?**
A likely explanation for the gain in MultiPV mode is that during
search each thread independently switches between rootMoves and via
the shared contempt score skews each other's evaluation.
**Why were the tests done with Skill Level=17?**
This was originally suggested by @Hanamuke and the idea is that with
Skill Level Stockfish sometimes plays also moves it thinks are slightly
sub-optimal and thus the quality of all moves offered by the MultiPV
search is checked by the test.
**Why are the ELO differences so huge?**
This is most likely because of the nature of Skill Level mode --
since it slower and weaker than normal mode, bugs in evaluation have
much greater effect.
---
Closes https://github.com/official-stockfish/Stockfish/pull/1515.
No functional change -- in single thread mode.
Use rootMoves[PVIdx].previousScore instead of bestValue for
dynamic contempt. This is equivalent for MultiPV=1 (bench remained the
same, even for higher depths), but more correct for MultiPV.
STC (MultiPV=3):
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 2657 W: 1079 L: 898 D: 680
http://tests.stockfishchess.org/tests/view/5aaa47cb0ebc590297330403
LTC (MultiPV=3):
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 2390 W: 874 L: 706 D: 810
http://tests.stockfishchess.org/tests/view/5aaa593a0ebc59029733040b
VLTC 240+2.4 (MultiPV=3):
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 2399 W: 861 L: 694 D: 844
http://tests.stockfishchess.org/tests/view/5aaf983e0ebc5902a182131f
LTC (MultiPV=4, Skill Level=17):
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 747 W: 333 L: 175 D: 239
http://tests.stockfishchess.org/tests/view/5aabccee0ebc5902997ff006
Note: although the ELO differences seem huge, they are inflated by the
nature of Skill Level / MultiPV search, so I don't think they can be
reasonably compared with classic ELO strength.
See https://github.com/official-stockfish/Stockfish/pull/1491 for some
verifications searches with MultiPV = 10 at depths 12 and 24 from the
starting position and the position after 1.e4, comparing the outputs
of the full PV by the old master and by this patch.
No functional change for MultiPV=1
This involves:
* replacing the union hacks with simply reusing the EntryPiece arrays
for the no-pawns case
* merging the PairsData structure with the EntryPiece/-Pawn structs
(with credit to Marco: @mcostalba)
* simplifying some HashTable functions
* thanks to previous changes, removing the ugly memsets
* simplifying the template logic for WDL/DTZ distinction
(now we distinguish based on an enum type, not the entry classes)
* removing the unneeded Atomic wrapper
-----------------------------
For reference, here is a manual way to check that patches concerning
table bases code are non-functional changes:
0) Download the Syzygy table bases (up to 6 men).
1) Make sure you have branches master and the pull request pointing to
the right commits.
2) Download the bench calculation scripts from the following URL:
https://gist.github.com/WOnder93/b5fcf9c989b4a1715684d5c82367cdbe
and copy into src inside your Stockfish repo.
3) Make the scripts executable (chmod +x *.sh).
4) Run the following command to use TBs located at <path>:
export SYZYGY_PATH='<path>'
5) After that, run this (it will take a long time, this is a deep bench):
BENCH_ARGS='128 1 22' ./check_benches.sh master tbprobe_cleanup 2>/dev/null`
==> You should see two equal numbers printed.
(Of course, now we have to trust that the script itself is correct :)
-----------------------------
Closes https://github.com/official-stockfish/Stockfish/pull/1477
No functional change.
