Master branch | |
Conda-forge |
Name | Downloads | Version | Platforms |
---|---|---|---|
MultiNEAT is a portable software library for performing neuroevolution, a form of machine learning that trains neural networks with a genetic algorithm. It is based on NEAT, an advanced method for evolving neural networks through complexification. The neural networks in NEAT begin evolution with very simple genomes which grow over successive generations. The individuals in the evolving population are grouped by similarity into species, and each of them can compete only with the individuals in the same species.
The combined effect of speciation, starting from the simplest initial structure and the correct matching of the genomes through marking genes with historical markings yields an algorithm which is proven to be very effective in many domains and benchmarks against other methods.
NEAT was developed around 2002 by Kenneth Stanley in the University of Texas at Austin.
GNU Lesser General Public License v3.0
http://multineat.com/docs.html
Prebuilt MultiNEAT package is available from conda-forge:
conda install multineat -c conda-forge
Conda-forge feedstock recipe can be found here.
Python 2.7 |
Python 3.5 |
Python 3.6 |
|
---|---|---|---|
Linux |
๐ | ๐ | ๐ |
macOS |
๐ | ๐ | ๐ |
Windows |
๐ | ๐ | ๐ |
Building MultiNEAT on Windows with python 2.7 is not possible, becuase it uses compiler from VS2008 that doesn't support C++11 features required by the library.
From now on only boost-python bindings are supported. So make sure to install boost and boost-python (e.g. from conda-forge) and as usual:
python setup.py build_ext
python setup.py install