by Jacob Zweig @ Northwestern University Visual perception, Neuroscience, and Cognition Lab
For decoding electrophysiologic signals using recurrent convolutional neural networks. Uses ensemble voting for classification. Mostly written around lasagne and scikit-learn. RCNNs are based on the architecture outlined in Recurrent Convolutional Neural Network for Object Recognition.
It's easiest to start with a scientific distribution. If you're in academia Enthought Canopy is free, and works well for me. People also like Anaconda. I'd recommend you install this in a virtual environment (see here), especially if you're on a shared computer, because I guarantee your labmates will mess up your python environment (sorry labmates, you're the best...).
First, cd to the RCNN_Toolbox folder and run the install command:
pip install -r requirements.txt
If that doesn't work, install these manually:
- numpy, scipy
- pyyaml
- Theano: installation instructions (Take special note of GPU installation instructions, because it will be really slow without it) Note: You should use the latest version of Theano, not the PyPI version. Install it with:
sudo pip install git+git://github.com/Theano/Theano.git
- Lasagne: (Install instructions)
- Optional but HIGHLY recommended: cuDNN (this will save you lots and lots of time, which you can use to watch Game of Thrones or learn a new way to tie your shoes)
I like to use Matlab to process my EEG data. If you do too, you might consider using the Matlab Engine for Python since it makes transferring data between the two programs quite seamless. To install the Matlab Engine for Python, follow the instructions here
Finally, to install this package, cd to the RCNN-Toolbox folder and run the install command:
sudo python setup.py install
If you're using cuDNN, enable CNMeM in your .theanorc file (I set it to .75 and it seems to work well).
In progress... see examples for an introduction to some of the functions in the toolbox