Stockfish is an open source UCI chess engine, developed by Tord Romstad, Joona Kiiski and Marco Costalba and licensed under the GNU General Public License version 3. The current version 4 (as of August 20, 2013) is available as C++ source code, and also has precompiled versions for Microsoft Windows, Mac OS X, and Linux 32-bit/64-bit. It can use up to thirty-two CPU cores in multiprocessor systems. The maximum size of the transposition tables is eight gigabytes. Stockfish implements an advanced alpha-beta search and uses bitboards.
This project modifies Stockfish and plugs in a neural evaluation function and a companion trainer for purely academic purposes. I thank Andrey Kotlarski without whose BPN, this project would be a lot harder to implement.
make profile-build ARCH=x86-64 (This is for 64-bit systems)
make profile-build ARCH=x86-32 (This is for 32-bit systems)
Go to 'Util' and extract the contents of the tarball. Then run the following commands,
sudo apt-get install tcl-dev
sudo apt-get install tk-dev
./configure
sudo make install
See the readme for rStockTrainer
- Open a terminal and navigate to the 'rStock' directory
- Type the command 'scid'
- In the GUI, go to 'Tools->Analysis engines->New'
- For the 'Name' option, give a suitable name
- For the 'Command' option, browse for the 'rStock' executable and select it
- Leave the other options to defalut
- Click on the 'Configure' button to configure the engine
Some points to remember while configuring the engine:
- There are 4 modes in the engine:
- 0 -> Original Stockfish with logging it's thinking lines
- 1 -> Online training
- 2 -> Play with an existing neural network
- 3 -> Original Stockfish
- For the training mode, the number of threads should be set to 1 (because the write to file cannot happen in parallel)
- The neural network file is usually set to 'stock.bpn'. Therefore make sure that any newly trained NN is named accordingly and placed in the rStock directory