ELO : 18479
Posts : 4398
Thanks received : 12001
on Fri 21 Mar 2014, 9:36 am by
ELO : 187
Posts : 46
Thanks received : 124
on Sat 22 Mar 2014, 9:40 am by
ELO : 14249
Posts : 4503
Thanks received : 8504
on Sat 22 Mar 2014, 5:45 pm by
ELO : 14249
Posts : 4503
Thanks received : 8504
on Sat 22 Mar 2014, 5:46 pm by
@Steve.R wrote:SPSA Tuner for Stockfish Chess Engine
SPSA Tuner is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
SPSA Tuner is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
1. Introduction to SPSA
Please see the following document (by James C. Spall) as an introduction to SPSA.
Unless otherwise stated, we follow the naming conventions and recommendations of the article.
2. Modifications for chess engine testing
Please see matlab pseudocode by Spall (document, p. 821) as a starting point.
For chess engine testing the following lines in the pseudocode
yplus = loss(thetaplus)
yminus = loss(thetaminus)
ghat = (yplus - yminus) / (2 * ck * delta)
are replaced with:
ghat = match(thetaplus, thetaminus) / (ck * delta)
match() plays a game between two identical engines.
If engine 1 (= thetaplus) wins, it returns 1.
If engine 2 (= thetaminus) wins, it returns -1.
If game is drawn, it returns 0.
For engine testing it makes sense to define a new varible "relative apply factor" Rk:
Rk := ak / ck^2.
It's interpretation is following:
theta := theta + ak * ghat
theta := theta + ak * match(thetaplus, thetaminus) / (ck * delta)
theta := theta + Rk * ck * match(thetaplus, thetaminus) / delta
Instead of trying to find ideal values for ak, we can try to find an ideal value for Rk.
3. Configuration file
To execute script 'spsa.pl', you need an INI style configuration file.
Take a look at example files provided. All options should be self-explanatory.
4. Variables CSV-file
Variables file (name of the file is defined in configuration file) is a comma separated (CSV) file.
Columns are defined as follows:
Column 0: Variable name (alphanumeric string)
Column 1: Variable initial value (float)
Column 2: Variable maximum value (float)
Column 3: Variable minimum value (float)
Column 4: Perturbation "ck" for the last iteration (float)
Column 5: Relative apply factor "Rk" for the last iteration (float)
Column 6: For simulation mode, this defines the ELO decrease for point x = (+/-) 100 compared to point x = 0 (optimum) (float).
- When ck is defined for the last iteration and the number of iterations is known, it's easy to derive a value for c.
- When ck and Rk are defined for the last iterations, it's easy to derive ak for the last iteration. Based on that we can derive value for a.
5. Getting Started
- Install Perl.
On Windows: Download and install Stawberry Perl (http://strawberryperl.com/).
On UNIX/Linux: It's part of the default installation. So you can skip this step.
- Install the following Perl Packages: Config::Tiny, Text::CSV, Math::Round.
On Windows: Open Command Prompt. Type: "cpan". Type: "install Config::Tiny". Type: "install Text::CSV". Type: "install Math::Round"
On Debian/Ubuntu Linux: Type: sudo apt-get install libconfig-tiny-perl libtext-csv-perl libmath-round-perl
On other UNIX/Linux systems: Please consult the documentation of relevant package management software.
- Modify stockfish to make parameters available via UCI. Copy the modified stockfish in the same directory with the script.
- Create a config file (just modify 'aggr.conf' slightly) and variable file (just modify 'aggr.var' and add more lines)
- Kick off the tuning by executing: spsa.pl [configFile]
6. Practical Guidelines
To be able to successfully use SPSA for tuning, one needs to determine good values for coefficients ck and Rk.
In most cases, it's impossible to determine ideal values.
For evaluation related parameters, we've had success with the following values:
Rk for the last iteration (CSV-file, column 5): 0.002
ck for the last iteration (CSV-file, column 4): 4 centipawns (= 8 in Stockfish's internal scale)
However if parameter is very insensitive ELO-wise, one needs to use a larger value for ck.
Also if one has even some sort of idea about the ELO-sensitivity of the parameter and how far from optimum
it might be at maximum, one can use the built-in simulator, to find good values for Rk and ck.
See files 'simul_1var.conf' and 'simul_1var.var' as an example.
ELO : 3284
Posts : 794
Thanks received : 2363
on Fri 22 Aug 2014, 4:18 pm by
50000/50000 games played
50000 @ 15+0.05 th 1
Tune queen mobility using SPSA. Low pri.
سبحان الله وبحمده سبحان الله العظيم