covers the most important cases from the user perspective:
It embeds the default net in the binary, so a download of that binary will result
in a working engine with the default net. The engine will be functional in the default mode
without any additional user action.
It allows non-default nets to be used, which will be looked for in up to
three directories (working directory, location of the binary, and optionally a specific default directory).
This mechanism is also kept for those developers that use MSVC,
the one compiler that doesn't have an easy mechanism for embedding data.
It is possible to disable embedding, and instead specify a specific directory, e.g. linux distros might want to use
CXXFLAGS="-DNNUE_EMBEDDING_OFF -DDEFAULT_NNUE_DIRECTORY=/usr/share/games/stockfish/" make -j ARCH=x86-64 profile-build
passed STC non-regression:
https://tests.stockfishchess.org/tests/view/5f4a581c150f0aef5f8ae03a
LLR: 2.95 (-2.94,2.94) {-1.25,-0.25}
Total: 66928 W: 7202 L: 7147 D: 52579
Ptnml(0-2): 291, 5309, 22211, 5360, 293
closes https://github.com/official-stockfish/Stockfish/pull/3070
fixes https://github.com/official-stockfish/Stockfish/issues/3030
No functional change.
This patch ports the efficiently updatable neural network (NNUE) evaluation to Stockfish.
Both the NNUE and the classical evaluations are available, and can be used to
assign a value to a position that is later used in alpha-beta (PVS) search to find the
best move. The classical evaluation computes this value as a function of various chess
concepts, handcrafted by experts, tested and tuned using fishtest. The NNUE evaluation
computes this value with a neural network based on basic inputs. The network is optimized
and trained on the evalutions of millions of positions at moderate search depth.
The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward.
It can be evaluated efficiently on CPUs, and exploits the fact that only parts
of the neural network need to be updated after a typical chess move.
[The nodchip repository](https://github.com/nodchip/Stockfish) provides additional
tools to train and develop the NNUE networks.
This patch is the result of contributions of various authors, from various communities,
including: nodchip, ynasu87, yaneurao (initial port and NNUE authors), domschl, FireFather,
rqs, xXH4CKST3RXx, tttak, zz4032, joergoster, mstembera, nguyenpham, erbsenzaehler,
dorzechowski, and vondele.
This new evaluation needed various changes to fishtest and the corresponding infrastructure,
for which tomtor, ppigazzini, noobpwnftw, daylen, and vondele are gratefully acknowledged.
The first networks have been provided by gekkehenker and sergiovieri, with the latter
net (nn-97f742aaefcd.nnue) being the current default.
The evaluation function can be selected at run time with the `Use NNUE` (true/false) UCI option,
provided the `EvalFile` option points the the network file (depending on the GUI, with full path).
The performance of the NNUE evaluation relative to the classical evaluation depends somewhat on
the hardware, and is expected to improve quickly, but is currently on > 80 Elo on fishtest:
60000 @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f28fe6ea5abc164f05e4c4c
ELO: 92.77 +-2.1 (95%) LOS: 100.0%
Total: 60000 W: 24193 L: 8543 D: 27264
Ptnml(0-2): 609, 3850, 9708, 10948, 4885
40000 @ 20+0.2 th 8
https://tests.stockfishchess.org/tests/view/5f290229a5abc164f05e4c58
ELO: 89.47 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 12756 L: 2677 D: 24567
Ptnml(0-2): 74, 1583, 8550, 7776, 2017
At the same time, the impact on the classical evaluation remains minimal, causing no significant
regression:
sprt @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f2906a2a5abc164f05e4c5b
LLR: 2.94 (-2.94,2.94) {-6.00,-4.00}
Total: 34936 W: 6502 L: 6825 D: 21609
Ptnml(0-2): 571, 4082, 8434, 3861, 520
sprt @ 60+0.6 th 1
https://tests.stockfishchess.org/tests/view/5f2906cfa5abc164f05e4c5d
LLR: 2.93 (-2.94,2.94) {-6.00,-4.00}
Total: 10088 W: 1232 L: 1265 D: 7591
Ptnml(0-2): 49, 914, 3170, 843, 68
The needed networks can be found at https://tests.stockfishchess.org/nns
It is recommended to use the default one as indicated by the `EvalFile` UCI option.
Guidelines for testing new nets can be found at
https://github.com/glinscott/fishtest/wiki/Creating-my-first-test#nnue-net-tests
Integration has been discussed in various issues:
https://github.com/official-stockfish/Stockfish/issues/2823https://github.com/official-stockfish/Stockfish/issues/2728
The integration branch will be closed after the merge:
https://github.com/official-stockfish/Stockfish/pull/2825https://github.com/official-stockfish/Stockfish/tree/nnue-player-wip
closes https://github.com/official-stockfish/Stockfish/pull/2912
This will be an exciting time for computer chess, looking forward to seeing the evolution of
this approach.
Bench: 4746616
A number of engines, GUIs and tournaments start to report WDL estimates
along or instead of scores. This patch enables reporting of those stats
in a more or less standard way (http://www.talkchess.com/forum3/viewtopic.php?t=72140)
The model this reporting uses is based on data derived from a few million fishtest LTC games,
given a score and a game ply, a win rate is provided that matches rather closely,
especially in the intermediate range [0.05, 0.95] that data. Some data is shown at
https://github.com/glinscott/fishtest/wiki/UsefulData#win-loss-draw-statistics-of-ltc-games-on-fishtest
Making the conversion game ply dependent is important for a good fit, and is in line
with experience that a +1 score in the early midgame is more likely a win than in the late endgame.
Even when enabled, the printing of the info causes no significant overhead.
Passed STC:
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 197112 W: 37226 L: 37347 D: 122539
Ptnml(0-2): 2591, 21025, 51464, 20866, 2610
https://tests.stockfishchess.org/tests/view/5ef79ef4f993893290cc146b
closes https://github.com/official-stockfish/Stockfish/pull/2778
No functional change
the option was, since at least 2014, not correctly implemented,
ignoring all dynamic adjustments to optimum time in search.
Instead of fixing it, remove it, no need to expose an option that
will influence time management negatively.
closes https://github.com/official-stockfish/Stockfish/pull/2765
No functional change.
Conceptually group hash clusters into super clusters of 256 clusters.
This scheme allows us to use hash sizes up to 32 TB
(= 2^32 super clusters = 2^40 clusters).
Use 48 bits of the Zobrist key to choose the cluster index. We use 8
extra bits to mitigate the quantization error for very large hashes when
scaling the hash key to cluster index.
The hash index computation is organized to be compatible with the existing
scheme for power-of-two hash sizes up to 128 GB.
Fixes https://github.com/official-stockfish/Stockfish/issues/1349
closes https://github.com/official-stockfish/Stockfish/pull/2722
Passed non-regression STC:
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 37976 W: 7336 L: 7211 D: 23429
Ptnml(0-2): 578, 4295, 9149, 4356, 610
https://tests.stockfishchess.org/tests/view/5edcbaaef29b40b0fc95abc5
No functional change.
This is a functional simplification of the time management system.
With this patch, there is a simple equation for each of two distinct
time controls: basetime + increment, and x moves in y seconds (+increment).
These equations are easy to plot and understand making future modifications
or adding additional time controls much easier.
SlowMover is reset to 100 so that is has no effect unless a user changes it.
There are two scaling variables:
* Opt_scale is a scale factor (or percentage) of time to use for this current move.
* Max_scale is a scale factor to apply to the resulting optimumTime.
There seems to be some elo gain in most scenarios.
Better performance is attributable to one of two things:
* minThinkingTime was not allowing reasonable time calculations for very short games like 10+0 or 10+0.01. This is because adding almost no increment and substracting move overhead for 50 moves quickly results in almost 0 time very early in the game. Master depended on minThinkingTime to handle these short games instead of good time management. This patch addresses this issue by lowering minThinkingTime to 0 and adjusting moverOverhead if there are very low increments.
* Notice that the time distribution curves tail downward for the first 10 moves or so. This causes less time to attribute for very early moves leaving more time available for middle moves where more important decisions happen.
Here is a summary of tests for this version at different time controls:
SMP 5+0.05
LLR: 2.97 (-2.94,2.94) {-1.50,0.50}
Total: 46544 W: 7175 L: 7089 D: 32280
Ptnml(0-2): 508, 4826, 12517, 4914, 507
https://tests.stockfishchess.org/tests/user/protonspring
STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 20480 W: 3872 L: 3718 D: 12890
Ptnml(0-2): 295, 2364, 4824, 2406, 351
https://tests.stockfishchess.org/tests/view/5ebc343e7dd5693aad4e6873
STC, sudden death
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 7024 W: 1706 L: 1489 D: 3829
Ptnml(0-2): 149, 813, 1417, 938, 195
https://tests.stockfishchess.org/tests/view/5ebc346f7dd5693aad4e6875
STC, TCEC style
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 4192 W: 1014 L: 811 D: 2367
Ptnml(0-2): 66, 446, 912, 563, 109
https://tests.stockfishchess.org/tests/view/5ebc34857dd5693aad4e6877
40/10
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 54032 W: 10592 L: 10480 D: 32960
Ptnml(0-2): 967, 6148, 12677, 6254, 970
https://tests.stockfishchess.org/tests/view/5ebc50597dd5693aad4e688d
LTC, sudden death
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 9152 W: 1391 L: 1263 D: 6498
Ptnml(0-2): 75, 888, 2526, 1008, 79
https://tests.stockfishchess.org/tests/view/5ebc6f5c7dd5693aad4e689b
LTC
LLR: 2.98 (-2.94,2.94) {-1.50,0.50}
Total: 12344 W: 1563 L: 1459 D: 9322
Ptnml(0-2): 70, 1103, 3740, 1171, 88
https://tests.stockfishchess.org/tests/view/5ebc6f4c7dd5693aad4e6899
closes https://github.com/official-stockfish/Stockfish/pull/2678
Bench: 4395562
This exploits the recent fractional Skill Level, and is a result from some discussion in #2221 and the older #758.
Basically, if UCI_LimitStrength is set, it will internally convert UCI_Elo to a matching fractional Skill Level.
The Elo estimate is based on games at TC 60+0.6, Hash 64Mb, 8moves_v3.pgn, rated with Ordo, anchored to goldfish1.13 (CCRL 40/4 ~2000).
Note that this is mostly about internal consistency, the anchoring to CCRL is a bit weak, e.g. within this tournament,
goldfish and sungorus only have a 200Elo difference, their rating difference on CCRL is 300Elo.
I propose that we continue to expose 'Skill Level' as an UCI option, for backwards compatibility.
The result of a tournament under those conditions are given by the following table, where the player name reflects the UCI_Elo.
# PLAYER : RATING ERROR POINTS PLAYED (%) CFS(%)
1 Elo2837 : 2792.2 50.8 536.5 711 75 100
2 Elo2745 : 2739.0 49.0 487.5 711 69 100
3 Elo2654 : 2666.4 49.2 418.0 711 59 100
4 Elo2562 : 2604.5 38.5 894.5 1383 65 100
5 Elo2471 : 2515.2 38.1 651.5 924 71 100
6 Elo2380 : 2365.9 35.4 478.5 924 52 100
7 Elo2289 : 2290.0 28.0 864.0 1596 54 100
8 sungorus1.4 : 2204.9 27.8 680.5 1596 43 60
9 Elo2197 : 2201.1 30.1 523.5 924 57 100
10 Elo2106 : 2103.8 24.5 730.5 1428 51 100
11 Elo2014 : 2030.5 30.3 377.5 756 50 98
12 goldfish1.13 : 2000.0 ---- 511.0 1428 36 100
13 Elo1923 : 1928.5 30.9 641.5 1260 51 100
14 Elo1831 : 1829.0 42.1 370.5 756 49 100
15 Elo1740 : 1738.3 42.9 277.5 756 37 100
16 Elo1649 : 1625.0 42.1 525.5 1260 42 100
17 Elo1558 : 1521.5 49.9 298.0 756 39 100
18 Elo1467 : 1471.3 51.3 246.5 756 33 100
19 Elo1375 : 1407.1 51.9 183.0 756 24 ---
It can be observed that all set Elos correspond within the error bars with the observed Ordo rating.
No functional change
Adding a clamp function makes some of these range limitations a bit prettier and removes some #include's.
STC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 28117 W: 6300 L: 6191 D: 15626
http://tests.stockfishchess.org/tests/view/5c9aa1df0ebc5925cfff8fcc
Non functional change.
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.
No functional change
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:
Contempt 21
ELO: 51.68 +-1.9 (95%) LOS: 100.0%
Total: 40000 W: 9487 L: 3581 D: 26932
http://tests.stockfishchess.org/tests/view/5bdb1a140ebc595e0ae2620a
Contempt 24
ELO: 52.21 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 9759 L: 3793 D: 26448
http://tests.stockfishchess.org/tests/view/5bdb1b680ebc595e0ae2620d
Bench: 3459874
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/
Mirrors:
http://tablebase.sesse.net/ (partial)
http://tablebase.lichess.ovh/tables/standard/7/
Generator code:
https://github.com/syzygy1/tb/
Closes https://github.com/official-stockfish/Stockfish/pull/1707
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).
Now Stockfish finds it instantly (as it should), nice work 👊👍 .
```
dep score nodes time
7 +132.79 4339 0:00.00 Rb1+ Kc4 Nd6+ Kc5 Bg1+ Kxd6 Rxb6+ Kc7 Be3 Rh2 Bd4
6 +132.79 1652 0:00.00 Rb1+ Kc4 Nd2+ Kd5 Rxb6 Rxh2 Nf3 Rf2
5 +132.79 589 0:00.00 Rb1+ Kc4 Rxb6 Rxh2 Nf6 Rh1+ Kb2
4 +132.79 308 0:00.00 Rb1+ Kc4 Nd6+ Kc3 Rxb6 Rxh2
3 +132.79 88 0:00.00 Rb1+ Ka4 Nc3+ Ka5 Ra1+ Kb4 Ra4+ Kxc3 Rxh4
2 +132.79 54 0:00.00 Rb1+ Ka4 Nc3+ Ka5 Ra1+ Kb4
1 +132.7
```
Change the operators of the Option type in uci.h to accept floating
point numbers in double precision on input as the numerical type for
the "spin" values of the UCI protocol.
The output of Stockfish after the "uci" command is unaffected.
This change is compatible with all the existing GUI (as they will
continue sending integers that we can interpret as doubles in SF),
and allows us to pass double parameters to Stockfish in the console
via the "setoption" command. This will be useful if we implement
another tuner as an alternative for SPSA.
Closes https://github.com/official-stockfish/Stockfish/pull/1556
No functional change.
---------------------
A example of the new functionality in action in the branch `tune_float2'`:
876c322d0f
I have added the following lines in ucioptions.cpp:
```C++
void on_pi(const Option& o)
{
double x = Options["PI"]; // or double x = o;
std::cerr << "received value is x = " << x << std::endl;
}
...
o["PI"] << Option(3.1415926, -10000000, 10000000, on_pi);
```
Then I can change the value of Pi in Stockfish via the command line, and
check that Stockfish understands a floating point:
````
> ./stockfish
> setoption name PI value 2.7182818284
received value is x = 2.71828
````
On output, the default value of Pi is truncated to 3 (to remain compatible
with the UCI protocol and GUIs):
````
> uci
[...]
option name SyzygyProbeLimit type spin default 6 min 0 max 6
option name PI type spin default 3 min -10000000 max 10000000
uciok
````
This patch introduces an Analysis Contempt UCI combo box to control
the behaviour of contempt during analysis. The possible values are
Both, Off, White, Black. Technically, the engine is supposed to be in
analysis mode if UCI_AnalyseMode is set by the graphical user interface
or if the user has chosen infinite analysis mode ("go infinite").
Credits: the idea for the combo box is due to Michel Van den Bergh.
No functional change (outside analysis mode).
-----------------------------------------------------
The so-called "contempt" is an optimism value that the engine adds
to one color to avoid simplifications and keep tension in the position
during its search. It was introduced in Stockfish 9 and seemed to give
good results during the TCEC 11 tournament (Stockfish seemed to play a
little bit more actively than in previous seasons).
The patch does not change the play during match or blitz play, but gives
more options for correspondance players to decide for which color(s) they
would like to use contempt in analysis mode (infinite time). Here is a
description of the various options:
* Both : in analysis mode, use the contempt for both players (alternating)
* Off : in analysis mode, use the contempt for none of the players
* White : in analysis mode, White will play actively, Black will play passively
* Black : in analysis mode, Black will play actively, White will play passively
To more clearly distinguish them from "const" local variables, this patch
defines compile-time local constants as constexpr. This is consistent with
the definition of PvNode as constexpr in search() and qsearch(). It also
makes the code more robust, since the compiler will now check that those
constants are indeed compile-time constants.
We can go even one step further and define all the evaluation and search
compile-time constants as constexpr.
In generate_castling() I replaced "K" with "step", since K was incorrectly
capitalised (in the Chess960 case).
In timeman.cpp I had to make the non-local constants MaxRatio and StealRatio
constepxr, since otherwise gcc would complain when calculating TMaxRatio and
TStealRatio. (Strangely, I did not have to make Is64Bit constexpr even though
it is used in ucioption.cpp in the calculation of constexpr MaxHashMB.)
I have renamed PieceCount to pieceCount in material.h, since the values of
the array are not compile-time constants.
Some compile-time constants in tbprobe.cpp were overlooked. Sides and MaxFile
are not compile-time constants, so were renamed to sides and maxFile.
Non-functional change.
Add a logarithmic term in the optimism computation, increase
the maximal optimism and lower the contempt offset.
This increases the dynamics of the optimism aspects, giving
a boost for balanced positions without skewing too much on
unbalanced positions (but this version will enter panic mode
faster than previous master when behind, trying to draw faster
when slightly behind). This helps, since optimism is in general
a good thing, for instance at LTC, but too high optimism
rapidly contaminates play.
passed STC:
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 159343 W: 34489 L: 33588 D: 91266
http://tests.stockfishchess.org/tests/view/5a8db9340ebc590297cc85b6
passed LTC:
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 47491 W: 7825 L: 7517 D: 32149
http://tests.stockfishchess.org/tests/view/5a9456a80ebc590297cc8a89
It must be mentioned that a version of the PR with contempt 0
did not pass STC [0,5]. The version in the patch, which uses
default contempt 12, was found to be as strong as current master
on different matches against SF7 and SF8, both at STC and LTC.
One drawback maybe is that it raises the draw rate in self-play
from 56% to 59%, giving a little bit less sensitivity for SF
developpers to find evaluation improvements by selfplay tests
in fishtest.
Possible further work:
• tune the values accurately, while keeping in mind the drawrate issue
• check whether it is possible to remove linear and offset term
• try to simplify the S-shape curve
Bench: 5934644
Using a SPSA tuning session to optimize the time management
parameters.
With SPSA tuning it is not always possible to say where improvements
came from. Maybe some variables changed randomly or because result
was not sensitive enough to them. So my explanation of changes will
not be necessarily correct, but here it is.
• When decrease of thinking time was added by Joost a few months ago
if best move has not changed for several plies, one more competing
indicator was introduced for the same purpose along with increase
in score and absence of fail low at root. It seems that tuning put
relatively more importance on that new indicator what allowed to save
time.
• Some of this saved time is distributed proportionally between all
moves and some more time were given to moves when score dropped a lot
or best move changed.
• It looks also that SPSA redistributed more time from the beginning to
later stages of game via other changes in variables - maybe because
contempt made game to last longer or for whatever reason.
All of this is just small tweaks here and there (a few percentages changes).
STC (10+0.1):
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 18970 W: 4268 L: 4029 D: 10673
http://tests.stockfishchess.org/tests/view/5a9291a40ebc590297cc8881
LTC (60+0.6):
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 72027 W: 12263 L: 11878 D: 47886
http://tests.stockfishchess.org/tests/view/5a92d7510ebc590297cc88ef
Additional non-regression tests at other time controls
Sudden death 60s:
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 14444 W: 2715 L: 2608 D: 9121
http://tests.stockfishchess.org/tests/view/5a9445850ebc590297cc8a65
40 moves repeating at LTC:
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 10309 W: 1880 L: 1759 D: 6670
http://tests.stockfishchess.org/tests/view/5a9566ec0ebc590297cc8be1
This is a functional patch only for time management, but the bench
does not reflect this because it uses fixed depth search, so the number
of nodes does not change during bench.
No functional change.
Make contempt dependent on the current score of the root position.
The idea is that we now use a linear formula like the following to decide
on the contempt to use during a search :
contempt = x + y * eval
where x is the base contempt set by the user in the "Contempt" UCI option,
and y * eval is the dynamic part which adapts itself to the estimation of
the evaluation of the root position returned by the search. In this patch,
we use x = 18 centipawns by default, and the y * eval correction can go
from -20 centipawns if the root eval is less than -2.0 pawns, up to +20
centipawns when the root eval is more than 2.0 pawns.
To summarize, the new contempt goes from -0.02 to 0.38 pawns, depending if
Stockfish is losing or winning, with an average value of 0.18 pawns by default.
STC:
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 110052 W: 24614 L: 23938 D: 61500
http://tests.stockfishchess.org/tests/view/5a72e6020ebc590f2c86ea20
LTC:
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 16470 W: 2896 L: 2705 D: 10869
http://tests.stockfishchess.org/tests/view/5a76c5b90ebc5902971a9830
A second match at LTC was organised against the current master:
ELO: 1.45 +-2.9 (95%) LOS: 84.0%
Total: 19369 W: 3350 L: 3269 D: 12750
http://tests.stockfishchess.org/tests/view/5a7acf980ebc5902971a9a2e
Finally, we checked that there is no apparent problem with multithreading,
despite the fact that some threads might have a slightly different contempt
level that the main thread.
Match of this version against master, both using 5 threads, time control 30+0.3:
ELO: 2.18 +-3.2 (95%) LOS: 90.8%
Total: 14840 W: 2502 L: 2409 D: 9929
http://tests.stockfishchess.org/tests/view/5a7bf3e80ebc5902971a9aa2
Include suggestions from Marco Costalba, Aram Tumanian, Ronald de Man, etc.
Bench: 5207156
Set the default contempt value of Stockfish to 20 centipawns.
The contempt feature of Stockfish tries to prevent the engine from
simplifying the position too quickly when it feels that it is very
slightly behind, instead keeping the tension a little bit longer.
Various tests in November 2017 have proved that our current imple-
mentation works well against SF7 (which is about 130 Elo weaker than
current master) and than the Elo gain is an increasing function of
contempt, going (against SF7) from +0 Elo when contempt is set at
zero centipawns, to +30 Elo when contempt is 40 centipawns.
See pull request 1325 for details:
https://github.com/official-stockfish/Stockfish/pull/1325
This november discussion left open the decision of which "default"
value for contempt we should use for Stockfish, taking into account
the various uses ofStockfish (opening preparation for humans, computer
online tournaments,analysis tool for web pages, human/computer play,
etc).
This pull request proposes to set the default contempt value of SF
to twenty centipawns, which turns out to be the highest value which
is not a regression against current master, as this seemed to be a
good compromise between risk and safety. A couple of SPRT[-3..1]
tests were done to bisect this value:
Contempt 10: http://tests.stockfishchess.org/tests/view/5a5d42d20ebc5902977e2901 (PASSED)
Contempt 15: http://tests.stockfishchess.org/tests/view/5a5d41740ebc5902977e28fa (PASSED)
Contempt 20: http://tests.stockfishchess.org/tests/view/5a5d42060ebc5902977e28fc (PASSED)
Contempt 25: http://tests.stockfishchess.org/tests/view/5a5d433f0ebc5902977e2904 (FAILED)
Surprisingly, a test at "very long time control" hinted that using
contempt 20 is not only be non-regressive against contempt 0, but
may actually exhibit some small Elo gain, giving a likehood of superio-
rity of 88.7% after 8500 games:
VLTC:
ELO: 2.28 +-3.7 (95%) LOS: 88.7%
Total: 8521 W: 1096 L: 1040 D: 6385
http://tests.stockfishchess.org/tests/view/5a60b2820ebc590297b9b7e0
Finally, there was some concerns that a contempt value of 20 would
be worse than a value of 7, but a test with 20000 games at STC was
neutral:
STC:
ELO: 0.45 +-3.1 (95%) LOS: 61.2%
Total: 20000 W: 4222 L: 4196 D: 11582
http://tests.stockfishchess.org/tests/view/5a64d2fd0ebc590297903868
See the comments in pull request 1361 for the long, nice discussion
(180 entries :-)) leading to the decision to propose contempt 20 as
the default value:
https://github.com/official-stockfish/Stockfish/pull/1361
Whether Stockfish should strictly adhere to the Komodo and Houdini
semantics and add the UCI commands to force the contempt to be White
in the so-called "analysis mode" is still under discussion, and may
be or may not be the object of a future commit.
Bench: 5783344
For efficiency reasons current master only allows for transposition table sizes that are N = 2^k in size, the index computation can be done efficiently as (hash % N) can be written instead as (hash & 2^k - 1). On a typical computer (with 4, 8... etc Gb of RAM), this implies roughly half the RAM is left unused in analysis.
This issue was mentioned on fishcooking by Mindbreaker:
http://tests.stockfishchess.org/tests/view/5a3587de0ebc590ccbb8be04
Recently a neat trick was proposed to map a hash into the range [0,N[ more efficiently than (hash % N) for general N, nearly as efficiently as (hash % 2^k):
https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/
namely computing (hash * N / 2^32) for 32 bit hashes. This patch implements this trick and now allows for general hash sizes. Note that for N = 2^k this just amounts to using a different subset of bits from the hash. Master will use the lower k bits, this trick will use the upper k bits (of the 32 bit hash).
There is no slowdown as measured with [-3, 1] test:
http://tests.stockfishchess.org/tests/view/5a3587de0ebc590ccbb8be04
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 128498 W: 23332 L: 23395 D: 81771
There are two (smaller) caveats:
1) the patch is implemented for a 32 bit hash (so that a 64 bit multiply can be used), this effectively limits the number of clusters that can be used to 2^32 or to 128Gb of transpostion table. That's a change in the maximum allowed TT size, which could bother those using 256Gb or more regularly.
2) Already in master, an excluded move is hashed into the position key in rather simple way, essentially only affecting the lower 16 bits of the key. This is OK in master, since bits 0-15 end up in the index, but not in the new scheme, which picks the higher bits. This is 'fixed' by shifting the excluded move a few bits up. Eventually a better hashing scheme seems wise.
Despite these two caveats, I think this is a nice improvement in usability.
Bench: 5346341
This shoudl reduce time losses experienced by
users after new time management code.
Verified for no regression in very short TC (4sec + 0.1)
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 35262 W: 7426 L: 7331 D: 20505
Bench 5322108
What this patch does is:
* increase safety margin from 40ms to 60ms. It's worth noting that the previous
code not only used 60ms incompressible safety margin, but also an additional
buffer of 30ms for each "move to go".
* remove a whart, integrating the extra 10ms in Move Overhead value instead.
Additionally, this ensures that optimumtime doesn't become bigger than maximum
time after maximum time has been artificially discounted by 10ms. So it keeps
the code more logical.
Tested at 3 different time controls:
Standard 10+0.1
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 58008 W: 10674 L: 10617 D: 36717
Sudden death 16+0
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 59664 W: 10945 L: 10891 D: 37828
Tournament 40/10
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 16371 W: 3092 L: 2963 D: 10316
bench: 5479946
Simplify out low level sync stuff (mutex
and friends) and avoid to use them directly
in many functions.
Also some renaming and better comment while
there.
No functional change.
a single Xeon Phi can present itself as a single numa node with up to 288 threads (4 threads per hardware core).
Tested to work as expected with a Xeon Phi CPU 7230 up to 256 threads.
No functional change
Closes#1045
Allow to specifiy the log file name, this comes
handy in case of self-matches so that each SF
instance writes into a different log file.
No functional change.
Simplify time management code by removing hard stops for unchanging first root moves.
Search is now stopped earlier at the end iteration if it did not have fail-lows at root.
This simplification also fixes pondering bug. Ponder flag was true by default
and cutechess-cli doesn't change it to false even though no pondering is possible.
Fix the issue by setting the default value of 'Ponder' flag to false.
10+0.1:
ELO: 3.51 +-3.0 (95%) LOS: 99.0%
Total: 20000 W: 3898 L: 3696 D: 12406
40+0.4:
ELO: 1.39 +-2.7 (95%) LOS: 84.7%
Total: 20000 W: 3104 L: 3024 D: 13872
60+0.06:
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 37231 W: 5333 L: 5236 D: 26662
Stopped run at 100+1:
LLR: 1.09 (-2.94,2.94) [-3.00,1.00]
Total: 37253 W: 4862 L: 4856 D: 27535
Resolves#523Fixes#510
The only interesting change is the moving of
stack[MAX_PLY+4] back to its original position
in id_loop (now renamed Thread::search).
No functional change.
Stéphane Nicolet