Code for reproducing the key results of our NIPS 2015 paper on semi-supervised low-rank logistic regression models for large functional neuroimaging datasets.

Bzdok D, Eickenberg M, Grisel O, Thirion B, Varoquaux G. Semi-supervised Factored Logistic Regression for High-Dimensional Neuroimaging Data Advances in Neural Information Processing Systems (NIPS 2015), Montreal. Paper on ResearchGate

Please cite this paper when using the code for your research.

To follow established conventions of scikit-learn estimators, the SSEncoder class exposes the functions fit(), predict(), and score(). This should allow for seamless integration into other scikit-learn-enabled machine-learning pipelines.

For questions and bug reports, please send me an e-mail at danilobzdok[at]gmail.com.

1. Make sure that recent versions of the following packages are available:

   • Python (version 2.7 or higher)
   • Numpy (e.g. pip install numpy)
   • Theano (e.g. pip install Theano)
   • Nilearn (e.g., pip install nilearn)
   • Nibabel (e.g., pip install nibabel)

2. Set floatX = float32 in the [global] section of Theano config (usually ~/.theanorc). Alternatively you could prepend THEANO_FLAGS=floatX=float32 to the python commands.

3. Clone this repository, e.g.:

git clone https://github.com/banilo/nips2015.git