The patch was tested for correctness by running bench with and
without the change against current master, and the tablebase hit
numbers were found to be identical in both cases. See the pull
request comments for details:
https://github.com/official-stockfish/Stockfish/pull/1826
31m059 [Sat, 24 Nov 2018 06:55:09 +0000 (01:55 -0500)]
Simplify casting extension
On November 16th, before the removal of the depth condition, I tried
revising castling extensions to only handle castling moves, rather than
moves that change castling rights generally. It appeared to be a slight
Elo gain at STC but insufficient to pass [0, 4] (+0.5 Elo), but what I
overlooked was that it made pos.can_castle(us) irrelevant and should
have been a simplification. Recent discussion with @Chess13234 and
Michael Chaly (@Vizvezdenec) inspired me to take a second look, and
the simplification continues to pass when rebased on the current master.
This replaces two conditions with one, because type_of(move) == CASTLING
implies pos.can_castle(Us), allowing us to remove the latter condition.
Turn on MADV_RANDOM for Syzygy mmaps (on Unix-like builds)
When running on a cloud VM (n1-highcpu-96) with several NVMe SSDs and
some non-SSDs for tablebases, I noticed that the average SSD request size was
more than 256 kB. This doesn't make a lot of sense for Syzygy tablebases,
which have a block size of 32 bytes and very low locality.
Seemingly, the tablebase access patterns during probing make the OS,
at least Linux, think that readahead is advantageous; normally, it
gives up doing readahead if there are too many misses, but it doesn't,
perhaps due to the fairly high overall hit rates. (It seems the kernel cannot
distinguish between reading a block that was paged in because the userspace
wanted it explicitly, and one that was read as part of readahead.)
Setting MADV_RANDOM effectively turns off readahead, which causes
the request size to drop to 4 kB. In the aforemented cloud VM test,
this roughly tripled the amount of I/O requests that were able to go
through, while reducing the total traffic from 2.8 GB/sec to 56 MB/sec
(moving the bottleneck to the non-SSDs; it seems the SSDs could have
sustained many more requests).
Jörg Oster [Sun, 25 Nov 2018 10:27:40 +0000 (11:27 +0100)]
Qsearch simplification. (#1828)
Don't do an extra TT update in case of a fail-high,
but simply break off the moves loop and let the TT update
at the end of qsearch do this job.
Same workflow/logic as in our main search function now.
Tested for no regression to be on the safe side.
STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 30237 W: 6665 L: 6560 D: 17012
http://tests.stockfishchess.org/tests/view/5bf928e80ebc5902bced3f3a
I've recently tried to introduce tropism difference term in kingdanger which
passed STC 6 times but failed LTC all the time. Maybe using quadratic scaling
for it will also be helpful.
31m059 [Sat, 24 Nov 2018 01:09:03 +0000 (02:09 +0100)]
Remove the tropism term from kingDanger
A recent LTC tuning session by @candirufish showed this term decreasing significantly. It appears that it can be removed altogether without significant Elo loss.
I also thank @GuardianRM, whose attempt to remove tropism from king danger inspired this one.
After this PR is merged, my next step will be to attempt to tune the coefficients of this new, simplified kingDanger calculation.
Because of aggressive time management and optimistic assumptions
about move overhead, it's still very easy to get Stockfish to forfeit
on time when we hit an endgame and have Syzygy EGTB on a spinning
drive. The latency from serving a few thousand EGTB probes (~10ms each),
of which there can currently be up to 4000 outstanding before a time
check, will easily overwhelm the default Move Overhead of 30ms.
This problem was first raised by Gian-Carlo Pascutto and some solutions
and improvements were discussed in the following pull requests:
https://github.com/official-stockfish/Stockfish/pull/1471
https://github.com/official-stockfish/Stockfish/pull/1623
https://github.com/official-stockfish/Stockfish/pull/1783
This patch is a minimal change proposed by Marco Costalba to lower
the impact of the bug. We now force a check of the clock right after
each tablebase read.
(1) The bonus value has not been carefully tested, so it may be possible
to find slightly better values.
(2) Plan is to now try adding similar restriction for pawns. I wanted to
include that as part of this pull request, but I was advised to do it as
two separate pull requests. STC is currently running here, but may not add
enough value to pass green.
Preparation commit for the upcoming Stockfish 10 version, giving a chance to catch last minute feature bugs and evaluation regression during the one-week code freeze period. Also changing the copyright dates to include 2019.
Personally, I feel like SF has been tuned to death recently and that we
need to step away from existing-parameter tunes for a bit and focus more
on new ideas. I don't really think there's much more ELO in these tunes
(for now). For me at least, this was the last existing-parameter tune I'll
be running for quite a while. Cheers!
protonspring [Fri, 2 Nov 2018 16:49:51 +0000 (10:49 -0600)]
Code style in search.cpp
It does not appear to be not necessary or advantageous to
conditionally initialize kingRing[Us] or kingAttackersCount[Them],
so the 'else' can be removed.
The values in the rook table now look a bit strange for a human eye
and are hard to explain, maybe it would be nice to simplify them
by hand and see if we can pass another (clean) double green with a
more regular array.
Vizvezdenec [Fri, 9 Nov 2018 01:02:24 +0000 (04:02 +0300)]
Change default contempt from 21 to 24 centipawns
To top the rating lists and get more interesting middle play, it
is a good habit to set the default contempt to the highest value
that does not regress against contempt=0. We recently decreased
PawnValueEg it is logical that to raise a little bit the default
higher contempt because of the following internal dependency in
line 334 of search.cpp :
````
int ct = int(Options["Contempt"]) * PawnValueEg / 100; // From centipawns
````
STC: contempt=24 passed non-regression vs contempt=0
http://tests.stockfishchess.org/tests/view/5bd6d7f80ebc595e0ae21e14
LTC: contempt=24 passed non-regression LTC vs contempt=0
http://tests.stockfishchess.org/tests/view/5bd6e0980ebc595e0ae21f07
On 2018-11-01, we also tested the effects of contempt=21 and contempt=24
against Stockfish 9, and the net result was neutral:
Nooby [Sat, 27 Oct 2018 20:12:34 +0000 (04:12 +0800)]
Clear TableBase mappings in Search::clear()
This patch will make possible to free mapped TB files with "ucinewgame" command.
We wrote this patch specifically to address a problem that arose while
running Stockfish with 7-piece tablebases as a kibitzer at TCEC for
extended periods of time across multiple games. It was noted that after
some time, the NPS of the kibitzing Stockfish (which is usually 3x faster
than the Stockfish actually competing) would drop precipitously, eventually
falling to preposterously low numbers until restarted.
Their eval bot basically inputs FEN, go infinite, stop and loop, it probably
didn't do ucinewgame either. As time goes it gradually slowed down and OS
starts to use swap, this is not reasonable since the engine only uses 16GB
hash and the machine has 1TB physical RAM and does nothing else.
mstembera [Sun, 11 Nov 2018 03:49:13 +0000 (19:49 -0800)]
Remove redundant king square parameter
We don't need to pass the king square as an explicit parameter to the functions
king_safety() and do_king_safety() since we already pass in the position.
31m059 [Sun, 11 Nov 2018 21:14:28 +0000 (16:14 -0500)]
Simplify tropism. (#1807)
We calculate tropism as a sum of two factors. The first is the number of squares in our kingFlank and Camp that are attacked by the enemy; the second is number of these squares that are attacked twice. Prior to this commit, we excluded squares we defended with pawns from this second value, but this appears unnecessary. (Doubly-attacked squares near our king are still dangerous.) The removal of this exclusion is a possible small Elo gain at STC (estimated +1.59) and almost exactly neutral at LTC (estimated +0.04).
Some small changes in evaluation to try to convince Stockfish to centralize
her rooks more in middle game and avoid trapping them in the corners. Joint
work by SFisGOD and snicolet.
The recently committed Fail-High patch (081af9080542a0d076a5482da37103a96ee15f64)
had a number of changes beyond adjusting the depth of search on fail high, with
some undesirable side effects.
1) Decreasing depth on PV output, confusing GUIs and players alike as described in
issue #1787. The depth printed is anyway a convention, let's consider adjustedDepth
an implementation detail, and continue to print rootDepth. Depth, nodes, time and
move quality all increase as we compute more. (fixing this output has no effect on
play).
2) Fixes go depth output (now based on rootDepth again, no effect on play), also
reported in issue #1787
3) The depth lastBestDepth is used to compute how long a move is stable, a new move
found during fail-high is incorrectly considered stable if based on adjustedDepth
instead of rootDepth (this changes time management). Reverting this passed STC
and LTC:
4) In the thread voting scheme, the rank of the fail-high thread is now artificially
low, incorrectly since the quality of the move is much better than what adjustedDepth
suggests (e.g. if it takes 10 iterations to find VALUE_KNOWN_WIN, it has very low
depth). Further evidence comes from a test that showed that the move of highest
depth is not better than that of the last PV (which is potentially of much lower
adjustedDepth).
I.e. this test http://tests.stockfishchess.org/tests/view/5bd37a120ebc595e0ae1e7c3
failed SPRT[0, 5]:
In a running 5+0.05 th 8 test (more than 10000 games) a positive Elo estimate is
shown (strong enough for a [-3,1], possibly not [0,4]):
http://tests.stockfishchess.org/tests/view/5bd421be0ebc595e0ae1f315
LLR: -0.13 (-2.94,2.94) [0.00,4.00]
Total: 13644 W: 2573 L: 2532 D: 8539
Elo 1.04 [-2.52,4.61] / LOS 71%
Thus, restore old behavior as a bugfix, keeping the core of the fail-high patch
idea as resolving scheme. This is non-functional for bench, but changes searches
via time management and in the threaded case.
ThreatByKing (tuned values)
Green STC (50.8k games)
http://tests.stockfishchess.org/tests/view/5bd1d5a00ebc595e0ae1cbec
LTC (I stopped this test at 71.2k games. It's likely yellow.)
http://tests.stockfishchess.org/tests/view/5bd263e70ebc595e0ae1d77e
WeakUnopposedPawn (tweak) by xoto (https://github.com/xoto10)
Green STC (102.8k games)
http://tests.stockfishchess.org/tests/view/5bd306bb0ebc595e0ae1e146
Yellow LTC (90.8k games)
http://tests.stockfishchess.org/tests/view/5bd3ea660ebc595e0ae1f16b
aspiTune1 (tweak) by vondele (https://github.com/vondele)
Green STC (125.9k games)
http://tests.stockfishchess.org/tests/view/5bd2ae100ebc595e0ae1dab0
Yellow LTC (107.9k games)
http://tests.stockfishchess.org/tests/view/5bd3eb700ebc595e0ae1f16f
Thank you @31m059 (Mark Tenzer) for helping me! Also, thank you very much
for recognizing my efforts. I genuinely appreciate it.
No-Regression test with 8 threads, tc=15+0.15:
http://tests.stockfishchess.org/tests/view/5bc874ca0ebc592439f85938
LLR: 2.94 (-2.94,2.94) [-3.00,1.00]
Total: 24740 W: 3977 L: 3863 D: 16900 Elo +1.60
This was a cooperation between me and Michael Stembera:
-me recognizing SF having problems with resolving FH's efficiently at
high depths, thus starting some tests based on consecutive FH's.
-mstembera picking up the idea with first success at STC & LTC (so full
credits to him!)
-me suggesting how to resolve the issues pinpointed by S.G on PR #1768
and finally restricting the logic to the main thread so that it don't
regresses at multi-thread.
Enable numa machinery only for STRICTLY MORE than 8 threads. Reason for this
change is that nowadays SMP tests are always done with 8 threads. That is a
problem for multi-socket Windows machines running on fishtest.
The patch adds a small random component (+-1) to VALUE_DRAW for the evaluation
of draw positions (mostly 3folds). This random component is not static, but
potentially different for each visit of the node (hence derived from the node
counter). The effect is that in positions with many 3fold draw lines, different
lines are followed at each iteration. This keeps the search much more dynamic,
as opposed to being locked to one particular 3fold.
An example of a position where master suffers from 3fold-blindness and this patch
solves quickly is the famous TCEC game 53:
FEN: 3r2k1/pr6/1p3q1p/5R2/3P3p/8/5RP1/3Q2K1 b - - 0 51
master doesn't see that this is a lost position (draw eval up to depth 50) as
Qf6-e6 d4-d5 (found by patch at depth 23) leads to a loss.
The 3fold-blindness is more important at longer TC, the patch was yellow STC and
LTC, but passed VLTC:
Currently we update (track up) the pv even in the fail high case.
However most times in such cases the pv in the ply below remains unset
because there we have value == alpha and so finally we see truncated
pv's (=just one move) in fail high cases.
Of course tracking down these pv's (+sending them to the gui) comes at a
certian cost, but no-regression tests passed:
N.B.: Digging also into qsearch was tried in another version but seemed
not to pass the tests. This means that we don't always will get a pv
until the very tips.
Miguel Lahoz [Thu, 20 Sep 2018 07:12:12 +0000 (15:12 +0800)]
Simplify evaluation of blockers_for_king
Currently, we have two evaluation terms which account for pinned pieces.
One is for all pinned pieces in kingDanger computation and another for
just pinned pawns in ThreatByRank. We can increase the relevant bonus
for kingDanger calculation and do away with the ThreatByRank, which
seems to just add more complexity.
ttCapture can be assigned to only once outside of the main loop. The patch seems
functional at higher depths (seems possible in the case of non-legal TTmoves that
are captures).
31m059 [Mon, 8 Oct 2018 18:43:45 +0000 (14:43 -0400)]
Combo
This PR is a combination of two unrelated [0, 4] patches that appeared promising
but not quite strong enough to pass on their own. The combination initially failed
STC with a positive score after a long run, and the subsequent speculative LTC test
passed.
* tweak_threatOnQueen4 :
Increase the middlegame components of ThreatByMinor[QUEEN]
and ThreatByRook[QUEEN] by 15 each. Bryan's (@crossbr) analysis of CCC Bonus Game 10
inspired several tests on penalizing a queen with limited safe mobility. While
attempting to implement this idea, I noticed that when I did not include the queen's
current square in the calculations, the Elo gains seemed to vanish--and only then did
I have the idea to revisit ThreatByMinor[QUEEN] and ThreatByRook[QUEEN], adding a
corresponding value to each. Without Bryan's work, this test would never have been
submitted. I would also like to recognize the efforts and contributions of @SFisGOD,
who also vigorously worked on this idea.
* Use pure static eval for null move pruning :
This idea was directly re-purposed from a promising test by Jerry Donald Watson
(@jerrydonaldwatson) in August. It was also independently developed and tested by
Stefan Geschwentner (@locutus2) previously.
Thank you all!
Note by snicolet: I use this non-functional change patch
as a pretext to correct the wrong bench number I introduced
in the message of the previous commit.
These two simplifications appear to be affecting and/or offsetting each other.
Neither can be removed independently, but in combination they pass -3,1.
Storing unconditionally the current generation and bound is equivalent to master.
Part of the condition was added as a speed optimization in #429.
Here the branch is fully eliminated.
The multithreaded test was run after a plausible suggestion by @mstembera that the effect of this could be larger with many cores. The result seems to indicate this doesn't really matter on the 8core architecture abundantly available on fishtest.
This patch is a combinaison of two parameters tweaks patches which
have failed as strong yellows at LTC recently, by Alain Savard (Rocky640)
and Fabian Fichter (ianfab):
http://tests.stockfishchess.org/tests/view/5b8a71e60ebc592cf2749b1d
http://tests.stockfishchess.org/tests/view/5b81ce3b0ebc5902bdbb6585
The null move reduction formula in line 769 of search.cpp is quite convoluted
and full of mysterious magic constants at the moment, it would certainly be
nice to simplify it and/or gain more Elo from it:
Rocky640 [Fri, 31 Aug 2018 13:30:16 +0000 (09:30 -0400)]
Long Diagonal Tweaks
a) Reduce PSQT values along the long diagonals on non-central squares
and increase the LongDiagonal bonus accordingly. The effect is to penalise
bishops on the long diagonal which can not "see" the 2 central squares.
The "good" bishops still have more or less the same bonus as current master.
b) For a bishop on a central square, because of the "| s" term in the code,
the LongDiagonalBonus was always given. So while being there, remove the "| s"
and compensate the central Bishop PSQT accordingly.
protonspring [Wed, 29 Aug 2018 00:48:45 +0000 (02:48 +0200)]
Remove PawnsOnBothFlanks
It looks like PawnsOnBothFlanks can be removed from initiative().
A barrage of tests seem to confirm that the adjustment to -110
does not gain elo to offset any potential loss by removing
PawnsOnBothFlanks.
The fact that endgames with all the pawns on only one flank are
drawish is a well-known chess idea, so it seems quite strange that
this can be removed so easily without losing Elo.
In the past there had been attempts to improve on PawnsOnBothFlanks
with similar concepts (for instance using the pawn span value), but
the tests were at best neutral. Maybe Stockfish is now mature enough
that these refined ideas would work to replace PawnsOnBothFlanks?
MJZ1977 [Wed, 29 Aug 2018 00:27:47 +0000 (02:27 +0200)]
Fix bug with "excludedMove" for probcut
Bugfix: "excludedMove" has to be skipped in the probcut loop too.
If it is not skipped, the probcut can exit quickly with a wrong return
value corresponding to the excluded move. See the following forum
thread for a discussion:
https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/GGithf_VwSU
Shrink the hash table of tablebases back to 4096 entries
There is no need to make this as large as 65536 just for the sake of the
single 7-man tablebase that happens to have the key 0xf9247fff. Idea for the
fix by Ronald de Man, who suggested simply to allow more buckets past the end.
We also implement Robin Hood hashing for the hash table, which takes the worst
-case search for full 7-man tablebases down from 68 to 11 probes (Also takes
the average probe length from 2.06 to 2.05). For a table with 8K entries, the
corresponding numbers would be worst-case from 9 to 4, with average from 1.30
to 1.29.
Ondrej Mosnacek [Tue, 28 Aug 2018 23:22:22 +0000 (01:22 +0200)]
Refactor pure static eval code
This commit tries to make the new pure static eval code more readable by
splitting up the nested assignments into separate lines and making a few
more cosmetic tweaks.
protonspring [Mon, 20 Aug 2018 16:29:46 +0000 (10:29 -0600)]
make DistanceRing more consistent
This is a non-functional change. By pre-incrementing minKingPawnDistance
instead of post-incrementing, we can remove this -1.
This also makes DistanceRing more consistent with the rest of stockfish
since it now holds an actual "distance" instead of a less natural distance-1.
In current master, PseudoAttacks[KING][ksq] == DistanceRingBB[ksq][0]
With this patch, it will be PseudoAttacks[KING][ksq] == DistanceRingBB[ksq][1]
ie squares at distance 1 from the king. This is more natural use of distance.
The current array size DistanceRingBB[SQUARE_NB][8] is still OK with the new
definition, because maximum distance between two squares on a chess board is
seven (for example Kh1 and a8).
I think that more elo can be found in tweaks of this parameters so I plan
to further try some "hand-tuning", including increasing/decreasing ratio of
two constants and making bonus assimetric to 0. Thx to @AndyGrant for helping
with github and @jerrydonaldwatson for original idea.
A recent commit introduced a decrease of the static evaluation of
an inner node dependent on the previous stat score, which finally
was also stored in the transposition table. Now only the unchanged
static evaluation are stored there.
Remark:
For the case that a static evaluation can be retrieved from the
transposition table the value is now used unchanged. Another test
which also applies the modification in this case failed:
http://tests.stockfishchess.org/tests/view/5b7af6df0ebc5902bdbae2f6
Follow-up for the previous patch: we use an affine formula to mix stats
and evaluation in search. The idea is to give a bonus if the previous
move of the opponent was historically bad, and a malus if the previous
move of the opponent was historically good.
More precisely, if x is the stat score of the previous move by the opponent,
we implement the following formulas to tweak the evaluation at an internal
node of the tree for our pruning decisions at this node:
if x = 0, use v' = eval(P)
if x > 0, use v' = eval(P) - 5 - x/1024
if x < 0, use v' = eval(P) + 5 - x/1024
For reference, the previous master had this simpler rule:
if x > 0, use v' = eval(P) - 10
if x <= 0, use v' = eval(P)
VoyagerOne [Thu, 16 Aug 2018 21:58:48 +0000 (23:58 +0200)]
Mix search stats with evaluation
Mix search stats with evaluation: if the opponent's move has a good historyStat,
then decrease the evaluation of the internal node a bit for the pruning decisions
during search.
• the use of the previous stat score can probably be simplified in lines 587 and 716
• we could try to use a continuous bonus based on the previous stat score, instead
of just a fixed offset of -10 when the opponent previous move was good.
----------
Comments by Stefan Geschwentner:
Interesting idea. Because only the eval in search is tweak this should only
influence the eval and static eval used at inner nodes, and not on the return
search value (which comes in the end from quiescence search), except through
saving in TT followed by a TT cutoff.
So essentialy this effects diverse pruning/reduction parts -- eval and static
eval are lowered for good opponent moves:
• tt cutoff (ttValue)
• improving (static eval)
• more razoring (eval)
• less futility pruning (eval)
• less null move pruning (eval + static eval) (but with little more depth)
• more probcut (static eval)
• more move futility pruning (static eval)
protonspring [Fri, 17 Aug 2018 08:19:32 +0000 (10:19 +0200)]
Simplify king file dependancy in evaluate_shelter()
Remove the special value we used for the file of the king in the
evaluate_shelter() function, and compensate by tweaking some of
the ShelterStrength[] array values.
We double in this patch the weight of the capture history table in the
local scoring of captures for move ordering.
The capture history table is indexed by the triplet (capturing piece,
capture square, captured piece) and gets information like "it seems to
have been historically good in that part of the search tree to capture
a pawn with a rook on g3, even if it seems to lose material", and affect
the normaly pure « Most Valuable Victim » ordering of captures.
Finished yellow at STC after 228842 games (posting a +1.36 Elo gain):
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 228842 W: 50894 L: 50152 D: 127796
http://tests.stockfishchess.org/tests/view/5b714bb00ebc5902bdba332d
Alain SAVARD [Tue, 14 Aug 2018 06:31:56 +0000 (08:31 +0200)]
Remove pawncount array in imbalance
This is a natural follow up to last commit where values on the
QuadraticOurs diagonal and some piece value deltas were changed.
@Stefano80 tried to simplify the newly introduced pawncount array
using QuadraticOurs[1][1] =52 and a -30 adjustment on pawn values
So I started a 80000 30+0.3 SPSA on the QuadraticOurs diagonal and
on the piece values using @Stefano80 start values.
SPSA gave the new values QuadraticOurs[1][1] =38 and a -33 on pawn
values (the other changes on QuadraticOurs were kept, but were not
ignificant according to this test
http://tests.stockfishchess.org/tests/view/5b710ccb0ebc5902bdba2f27)
GuardianRM [Sun, 12 Aug 2018 16:40:03 +0000 (18:40 +0200)]
Non-linear bonus for pawn count
This patch introduces a non-linear bonus for pawns, along with some
(linear) corrections for the other pieces types.
The original values were obtained by a massive non-linear tuning of both
pawns and other pieces by GuardianRM, while Alain Savard and Chris Cain
later simplified the patch by observing that, apart from the pawn case, the
tuned corrections were in fact almost affine and could be incorporated in
our current code base via the piece values in types.h (offset) and the diagonal
of the quadratic matrix (slope). See discussion in PR#1725 :
https://github.com/official-stockfish/Stockfish/pull/1725
How to continue from there?
- Maybe the non-linearity for the pawn value could be somewhat tempered
again and a simpler linear correction for pawns would work?
Original patches:
- [Piece values](http://tests.stockfishchess.org/tests/view/5b6d2cc00ebc5902bdba02d5) by Stefano Cardanobile
- [Stat bonus](http://tests.stockfishchess.org/tests/view/5b6adbc90ebc5902bdb9da73) by Stefan Geschwentner
- [Rook on pawn](http://tests.stockfishchess.org/tests/view/5b62a95b0ebc5902bdb961c0) by Mark Tenzer
- [Hanging bonus](http://tests.stockfishchess.org/tests/view/5b5d2fa00ebc5902bdb90855) by Ivan Ilvec
- [ss tweak](http://tests.stockfishchess.org/tests/view/5b58b7240ebc5902bdb89025) by miguel-l
Simple razoring: depth 1 only, no distinction between PV / NonPV
We simplify the razoring logic by applying it to all nodes at depth 1 only.
An added advantage is that only one razor margin is needed now, and we treat
PV and Non-PV nodes in the same manner.
How to continue?
- There may be some conditions in which depth 2 razoring is beneficial.
- We can see whether the razor margin can be tuned, perhaps even with a
different value for PV nodes.
- Perhaps we can unify the treatment of PV and Non-PV nodes in other parts
of the search as well.
We are opting for this patch as the more logical and simple of the two,
and it appears to be no less strong. Thanks in particular to @DU-jdto
for input into this patch.
Miguel Lahoz [Thu, 9 Aug 2018 14:04:36 +0000 (22:04 +0800)]
Remove Condition For Passed Pawns
Currently, we do not consider pawns passed if there is another pawn of
the same color in front of them. It appears that this condition is not
necessary. The idea is that the doubled pawns are likely to be weak and
one of them will be likely captured anyway. On the other hand, if we do
somehow manage to promote a pawn, then the pawn behind it becomes passed
as well. In any case, the end result is we end up with an extra
potentially passed pawn. The current evaluation for passed pawns already
handles this case by also scaling down this effect.
Unify the "quiet" and "non-quiet" reduction rules for use at any kind of moves.
The idea behind it was that both rules reduce at similiar cases in master:
one directly for late previous moves and the other indirectly by using a
bad stat score which is used for most move sorting and so approximates the
late move condition.
For captures/promotions the old rule was triggered in 25% but the new
rule only for 3% of all cases (so now more reductions are done, whereas
for quiet moves reductions keep the same level).
Currently, we first calculate some bitboards at the top of Evaluation::space()
and then check whether we actually need them. Invert the ordering. Of course this
does not make a difference in current master because the constexpr bitboard
calculations are in fact done at compile time by any decent compiler, but I find
my version a bit healthier since it will always meet or exceed current implementation
even if we eventually change the spaceMask to something not contsexpr.
FauziAkram [Wed, 8 Aug 2018 15:48:06 +0000 (17:48 +0200)]
King Psqt Tuning
After a session of tuning for King Psqt I got some new values, which was later
tweaked manually by me Fauzi, to result in an Elo-gain patch which seems to scale
pretty well:
Introduce voting system for best move selction in multi-threads mode.
Joint work with Stefan Geschwentner, based on ideas introduced by
Michael Stembera.
Moves are upvoted by every thread using the margin to the minimum score
across threads and the completed depth.
First thread voting for the winner move is selected as best thread.
Passed STC, LTC. A further LTC test with only 4 threads failed with positive
score. A LTC with 31 threads was stopped with LLR 0.77 after 25k games to
avoid use of excessive resources (equivalent to 1.5M STC games).
Similar ideas were proposed by Michael Stembera 2 years ago #507, #508.
This implementation seems simpler and more understandable, the results
slightly more promising.
Further possible work:
1) Tweak of the formula using for assigning votes.
2) Use a different baseline for the score dependent part: maximum score
or winning probability could make more sense.
3) Assign votes in `Thread::Search` as iterations are completed and use
voting results to stop search.
4) Select best thread as the threads voting for best move with the highest
completed depth or, alternatively, vote on PV moves.
Marco Costalba [Sat, 28 Jul 2018 13:33:39 +0000 (15:33 +0200)]
Improve Stats definition
Use operator const T&() instead of operator T() to avoid possible
costly hidden copies of non-scalar nested types.
Currently StatsEntry has a single member T, so assuming
sizeof(StatsEntry) == sizeof(T) it happens to work, but it's
better to use the size of the proper entry type in std::fill.
Note that current code works because std::array items are ensured
to be allocated in contiguous memory and there is no padding among
nested arrays. The latter condition does not seem to be strictly
enforced by the standard, so be careful here.
Finally use address-of operator instead of get() to fully hide the
wrapper class StatsEntry at calling sites. For completness add
the arrow operator too and simplify the C++ code a bit more.
Same binary code as previous master under the Clang compiler.
This is the first patch teaching Stockfish how to use the 7-pieces
Syzygy tablebase currently calculated by Bujun Guo (@noobpwnftw) and
Ronald de Man (@syzygy1). The 7-pieces database are so big that they
required a change in the internal format of the files (technically,
some DTZ values are 16 bits long, so this had to be stored as wide
integers in the Huffman tree).
Here are the estimated file size for the 7-pieces Syzygy files,
compared to the 151G of the 6-pieces Syzygy:
```
7.1T ./7men_testing/4v3_pawnful (ongoing, 120 of 325 sets remaining)
2.4T ./7men_testing/4v3_pawnless
2.3T ./7men_testing/5v2_pawnful
660G ./7men_testing/5v2_pawnless
117G ./7men_testing/6v1_pawnful
87G ./7men_testing/6v1_pawnless
```
Some pointers to download or recalculate the tables:
Location of original files, by Bujun Guo:
ftp://ftp.chessdb.cn/pub/syzygy/
Bench: 5591925 (No functional change if SyzygyTB is not used)
----------------------
Comment by Leonardo Ljubičić (@DragonMist)
This is an amazing achievement, generating and being able to use 7 men syzygy
on the fly. Thank you for your efforts @noobpwnftw !! Looking forward how this
will work in real life, and expecting some trade off between gaining perfect
play and slow disc Access, but once the disc speed and space is not a problem,
I expect 7 men to yield something like 30 elo at least.
-----------------------
Comment by Michael Byrne (@MichaelB7)
This definitely has a bright future. I turned off the 50 move rule (ala ICCF
new rules) for the following position: `[d]8/8/1b6/8/4N2r/1k6/7B/R1K5 w - - 0 1`
This position is a 451 ply win for white (sans the 50 move rule, this position
was identified by the generator as the longest cursed win for white in KRBN v KRB).
This patch adds the tropism measure as a new term in the king danger variable.
Since we then trasform this variable as a Score via a quadratic formula, the
main effect of the patch is the positive correlation of the tropism measure
with some checks and pins information already present in the king danger code.
Miguel Lahoz [Fri, 27 Jul 2018 16:23:53 +0000 (00:23 +0800)]
Increase the mg->eg gradient for the PawnlessFlank malus
Just a change of value to S(19, 84). Also somewhat of a follow up
to the recent tweak in definition of KingFlank.
I tried a lot of other values before this, increasing and decreasing
but with little success, and before giving up I wanted to try tweaking
the middlegame and endgame values in the opposite directions. I guess
this is somewhat lucky.
The previous commit wouldn't compile on the Microsoft Virtual Studio C++ compiler. So use a more compatible style for the same idea (which we already use in numerous places of evaluate.cpp, for instance in line 563).
Under the Clang compiler, both versions generate exactly the same machine code (same md5 signatures for the two binaries).
Remove a popcount for HinderPassedPawn, and compensate by doubling
the bonus from S(4,0) to to S(8,0).
Maybe it was pure luck, but we got the idea of this Elo gaining patch by
seing the simplification attempt by Mike Whiteley in pull request #1703.
This suggests that whenever we have a passed evaluation simplification,
we should consider the possibility that the master bonus has become
slightly out of tune with time, and we should try a few Elo gaining [0..4]
tests by hand-tuning the master bonus.
Miguel Lahoz [Wed, 25 Jul 2018 16:11:51 +0000 (00:11 +0800)]
Tweak KingFlank when king is on edge files
This tweak excludes files D and E from the KingFlank bitboard when our
king is on the A or H files respectively. As far as I can tell, this
affects two things: the calculation for CloseEnemies and PawnlessFlank.
Aside from filtering out slightly less relevant attacks in the flank,
I suspect this helps with king prophylaxis, avoiding attacks and moving
towards the center when the pawns start to come off.
How to continue: Tweaking the two bonuses mentioned might give some
gain, although as far as I can tell, CloseEnemies is very sensitive to
even small changes.
I don't measure any speed-up on my system, with two parallel benches at depth 22:
Total time (ms) : 74989
Nodes searched : 144830258
Nodes/second : 1931353
master
Total time (ms) : 75341
Nodes searched : 144830258
Nodes/second : 1922329
testedpatch
And anyway, like Stefan Geschwentner, I don't think that a 0.3% speed-up would
be enough to pass a [0..5] LTC test -- as a first approximation, we have this
rule of thumb that 1% speed-up gives about 1 Elo point.
However, considering the facts that the reformatting by itself is interesting,
that this is your first green test and that you played by the rules by running
the SPRT[0..5] test before opening the pull request, I will commit the change.
I will only take the liberty to change the occurrences of safe in lines 590 and
591 to b, to make the code more similar to lines 584 and 585.
How to continue after this patch: there is now a slight semantic
overlap between the ThreatByPawnPush and the ThreatBySafePawn bonuses,
so hand-tuning either of these, or both at the same time, is natural.
This patch implements some idea by Alain Savard and Mike Whiteley taken from the perpertual renaming/reformatting thread.
This is a pure code cleaning patch (so no change in functionality), but I use it as a pretext to correct the bogus bench number that I introduced in the previous commit.
2) So there was a suspicion that the original fast passing LTC test which lead
us to accept the patch may have been a statistical accident, so we organized
a match against the previous master at LTC to get an Elo estimate for the
patch:
Replace the depth part in the reduction formula for higher depths
with a slower growing linear function. So for depth > 3 less reductions
are used.
What we can try next:
- move the break point to even higher depths
- tweak the slope for lower and higher depth
- even possibly use a further higher depth threshold for a another
slower growing function
Remove stronglyProtected Queen for ThreatByRook. Idea is that in the
current master the SliderOnQueen bonus and the see_ge() function do
something similar as ThreatByRook for Queen, so this patch removes
some redundancy, in that sense.
This is a non-functional simplification. We change replaces an 'OR'
and a lookup (rank_bb(ksq)) with a bitwise ~. This is fewer operations
and is probably faster.
This is a follow-up of the previous pull request (#1686) by Miguel.
We simplify the "Overload" bonus condition by re-using the "weak"
variable, which captures well the essence of the overload condition.
This may also be a small speed optimization because the weak variable
is in a register at this point of the code.
Miguel Lahoz [Thu, 19 Jul 2018 11:18:30 +0000 (19:18 +0800)]
Remove connectivity.
There seems to be some strange interaction between Overload and Connectivity.
Overload encourages us to not have too many defended and attacked pieces,
as this may expose us to various tactics. This feels somewhat like it is in
conflict with Connectivity, where pieces are defended preemptively.
Here I take the "pick one or the other" approach and just remove connectivity,
while strengthening the effect of Overload to compensate. The reasoning is that
if we defend our pieces preemptively, then it does get attacked, we want to do
something about it so we don't get penalized by Overload. On the other
hand, if it doesn't get attacked, then there's no need to defend it.
Use in addition the counter move history table for sorting quiet
check evasion moves in main and quiecence search. Also rename
"contHistory" to "continuationHistory" while there.
Miguel Lahoz [Mon, 16 Jul 2018 16:51:43 +0000 (00:51 +0800)]
Use single value for KingProtector.
After some recent big tuning session, the values for King Protector were
simplified to only be used on minor pieces. This patch tries to further
simplify by just using a single value, since current S(6,5) and S(5,6)
are close to each other. The value S(6,6) ended up passing, although
S(5,5) was also tried and failed STC.
Extend the bonus for Overload to cases where our side
has more than one attacker to a non pawn piece.
Based on an idea by Bryan in the forum. For instance,
now black gets the overload bonus in this position:
8/5R1k/6pb/p6p/P1N4P/1Pp5/2K3P1/2N4r b - - 6 46
because two black pieces are attacking the knight on c1
that is defended only by the king.