This project is mirrored to this personal github for demonstration purposes only and should not be used for research outside of the Redish lab. Please contact me at adkin099@umn.edu for more information or if you want to use the project.
Machine learning for neuroscience
An extension of scikit-learn with a focus on:
- verbosity: clear and extensive logging
- multisignal analysis: combine data from multiple sensors leveraging the
- noise handling: filter noisy data and predictions from neural signals, remove bad data from the training set
- cross validation: decoding should be cross-validated to ensure separation of training and test sets
- timestamped data: neural data is a temporal process (support could be better thusfar)
Please note, Python 2 is not supported and Python 3 is required.
Installation can be done easily via pip
git clone https://github.umn.edu/RedishLab/mlneuro.git
cd mlneuro
pip install -e .
On Windows, if MiniConda/Anaconda is installed, an environment file is included to install dependencies
git clone https://github.umn.edu/RedishLab/mlneuro.git
cd mlneuro
conda env create -n mlneuro -f environment.yml
The mlneuro environment may then be launched with activate mlneuro. If all optional dependencies are desired, specify the environment_full.yml file instead
To install mlneuro within the conda environment, activate the environment and use pip from the mlneuro root directory
activate mlneuro
pip install -e .
Dependencies should be installed by pip by default, but core dependencies include
- sklearn (requires scipy, numpy)
- numba
Additional optional dependencies for full-functionality include
- bufferkdtree** (GPU powered KDE with k-nearest neighbors)
- keras (neural network based estimators)
- statsmodels (poisson GLM estimator)
- h5py (saving/loading from hdf5 files)
- matplotlib (visuals and example plots)
Optional dependencies will not be auto-installed but can be found using pip or added via
conda env update -n mlneuro -f environment_full.yml
** The bufferkdtree requires POSIX-C headers, OpenCL, GPU Drivers, and Swig and
is not easily installed by conda and is consequently not included in environment_full.yml.
See the documentatino for the package if more details on installation are needed.
Example datasets require git large file support which is not enabled on the UofM github instance yet. The data folder will be available in the Redish lab team drive and
must be placed in mlneuro/mlneuro/datasets/ for example loading to work.
Documentation building requires sphinx, sphinx-gallery, sphinx_rtd_theme and matplotlib
In the head directory, make docs will build the documentation in
docs/build. Building documentation with the examples may take a long time, to generate documentation without example plots run make docs-no-plot
Windows does not support make or making without plots (yet), but sphinx generates a batch file for basic building which can be found in the docs folder.
cd docs
./make.bat