Installation¶
Release version¶
Run pip install snntoolbox
. This will install the minimum dependencies
needed to get started. Optional dependencies like matplotlib and scipy enable
generating output plots.
The toolbox relies on Keras internally, and by default installs Tensorflow as Keras backend. You may want to update this backend to optimally fit your system.
Note
The SNN toolbox provides a built-in simulator to run the converted network. This simulator is Keras-based and will use either Tensorflow or Theano as backend. Depending on the backend you choose, different features are available in the toolbox simulator. You can install both backends and switch between them simply by setting the corresponding parameter in the config file:
[simulation]
keras_backend = tensorflow
As the Theano is not actively developed any more, the tensorflow backend is better maintained. The only reason to choose Theano at this point is that we provide an implementation of MaxPooling in INIsim, which is not available with the tensorflow backend.
Development version (recommended)¶
To get the latest updates, checkout the repository.
In the toolbox root directory snn_toolbox/
, run pip install .
.
Note
Using easy_install via python setup.py install
has been reported
to fail on some platforms due to dependency issues.
Additional tools¶
For testing a converted network, the toolbox includes a ready-to-use spiking simulator. In addition, you may install and use one of the simulators described here.
Note
Depending on the simulator you use, we recommend installing the toolbox in a virtual environment, because simulators supported by pyNN may require different versions of their dependencies (Brian for instance only works with python2).