Installation

You can install using pip (Windows, MacOSX, Linux), binary packages or from source.

pip (Windows, MacOSX, Linux)

MDP is listed in the Python Package Index and can be installed with pip:

pip install mdp

Binary packages (Linux)

Users of some Linux distributions (Debian, Ubuntu, Fedora, …) can install the python3-mdp package using the distribution package manager.

Install the development version

If you want to live on the bleeding edge, check out the MDP git repositories. You can either browse the repository or directly install the development version from the repository with:

pip install git+https://github.com/mdp-toolkit/mdp-toolkit.git

Optional Libraries

MDP can make use of several additional libraries if they are installed on your system. They are not required for using MDP, but may give more functionality. Here a list of optional libraries and the corresponding additional features in MDP:

  • SciPy : Use the fast and efficient LAPACK wrapper for the symmetrical eigensolver, used interally by many nodes; use the fast FFT routines in some nodes; provide the Convolution2DNode, using the fast convolution routines in SciPy.
  • LibSVM : provide the LibSVMClassifier node.
  • joblib : provide the caching extension and the corresponding cache context manager.
  • scikit-learn : provide wrapper nodes to several sklearn algorithms.

You can install all the additional libraries with pip:

pip install scipy scikit-learn joblib libsvm

Testing

If you have successfully installed MDP, you can test your installation in a Python shell as follows:

>>> import mdp
>>> mdp.test()
>>> import bimdp
>>> bimdp.test()

Note that you will need to install pytest to run the tests.

If some test fails, please file a bug report. Optionally, report it to the mailing list.

License

MDP is distributed under the open source BSD license.