Compiling and Running - xcsf-dev/xcsf GitHub Wiki

PyPI

PyPI package Python versions

Some compiled wheels are available within the pip package manager, as well as a source distribution.

$ pip install --upgrade pip
$ pip install wheel
$ pip install xcsf

If installed in this way, the examples and notebooks will need to be copied from the repository.

If this fails to work, see the instructions below.

Building

Requirements

Stand-alone binary:

Python library:

  • All of the above for building the stand-alone executable.
  • C++11 compliant compiler.
  • Python (>= 3.6)

Compiler Options

  • XCSF_MAIN=ON : Build the stand-alone main executable (CMake default = ON)
  • XCSF_PYLIB=ON : Build the Python library (CMake default = OFF)
  • PARALLEL=ON : CPU parallelised matching, predicting, and updating with OpenMP (CMake default = ON)
  • ENABLE_TESTS=ON : Build and execute unit tests (CMake default = OFF)
  • NATIVE_OPT=ON : Optimise for the native system architecture (CMake default = ON)
  • SANITIZE=ON : Enable AddressSanitizer and UndefinedBehaviorSanitizer - compile as Debug (CMake default = OFF)

Ubuntu

$ sudo apt install python3 python3-dev cmake
$ git clone --recurse-submodules https://github.com/xcsf-dev/xcsf.git
$ cd xcsf/build
$ cmake -DCMAKE_BUILD_TYPE=Release -DXCSF_PYLIB=ON -DENABLE_TESTS=ON ..
$ make

macOS

Note: Try GCC if having problems with AppleClang and OpenMP.

$ brew install libomp cmake python
$ git clone --recurse-submodules https://github.com/xcsf-dev/xcsf.git
$ cd xcsf/build
$ cmake -DCMAKE_BUILD_TYPE=Release -DXCSF_PYLIB=ON -DENABLE_TESTS=ON ..
$ make

Windows

Git + CMake + MinGW64-gcc-8.1.0 + Python 3.10.0 x86-64

Notes: use 64-bit versions for both Python and mingw. Tested with: python-3.10.0-amd64.exe and mingw-w64-install.exe (8.1.0; x86_64; posix; seh; 0). A simple method to get started is: Start → MinGW-W64 project → Run terminal → change to the desired install location and enter the commands below. [Compilation should also be possible within an IDE such as Visual Studio or Eclipse.]

$ git clone --recurse-submodules https://github.com/xcsf-dev/xcsf.git
$ cd xcsf\build
$ cmake -DCMAKE_BUILD_TYPE=Release -DXCSF_PYLIB=ON -DENABLE_TESTS=ON -G "MinGW Makefiles" ..
$ cmake --build . --config Release

Documentation

Doxygen + graphviz

After running CMake:

$ make doc

Alternatively see: XCSF documentation.

Running

Stand-alone

Default parameters are hard-coded. These are overridden by default.json at runtime. The JSON configuration file follows a similar structure to the Python library. Currently there are 3 built-in problem environments: {csv, mp, maze}.

Example real-multiplexer classification:

$ ./xcsf/main mp 6

Example discrete mazes:

$ ./xcsf/main maze ../env/maze/maze4.txt

Example regression: learning env/csv/sine_3var_train.csv and testing env/csv/sine_3var_test.csv

$ ./xcsf/main csv ../env/csv/sine_3var

Python

After building with CMake option: -DXCSF_PYLIB=ON

See the Python examples and notebooks.

For example, single-step reinforcement learning:

$ python3 example_rmux.py

Note that many of the examples require a few other Python packages, which can be installed:

$ pip install numpy matplotlib tqdm sklearn gym

See Python Library Usage for more information.