- updated to accomodate pthon 3.11.9 and mne-python 1.7.0
- Improved LSTM architecture
- Re-implementation of the Brent-optimization method for source scaling instead of the RMS method.
- increased default source extent from 1-40 mm to 1-50 mm
- reduced default target_snr from 2-20 to 1-20
- changed default beta exponent from 0.5-1.5 to 0.5-3 to allow for smoother source time courses overall
- source_spread can now be mixed (region growing AND spherical expansion)
- introduced source_number_weighting:
weight = 1/number_of_sources
i.e., the sampling weight is the inverse of the number of sources. Example: If you desire between 1 and 5 sources, the change of having a single source is 48 %, the chance of having 5 sources is 12 %.
- Implemented the reverting of the time dimension of training data as data augmentation. This improved validation loss in our tests.
- Introducing region growing for source simulations. Region growing allows for more realistic source simulations that do not extend spherically but along a graph of connected surface dipoles. See esinet. Simulation.settings for details.
- Minor changes which were forgotten.
- Support for LSTM, Fully-Connected and ConvDip models
- Most recent code basis which served for the LSTM bioRxiv preprint
- Changed API from functional to object-oriented
- New Tutorials
- Models can now train on either single time instances of EEG data or on temporal data (LSTM Model)
- Optimized simulation through more intelligent matrix multiplications
- Im am sorry if this update broke your code. If you experience issues please either follow the tutorials on the new API or revert back to an older version.
- Made tutorial notebook clearer and minor changes
- Changed dependencies to be more rigid
- Changed progress bar to juypter mode
- added option to have mixed gaussian and flat source shapes for higher simulation diversity
- minor changes
- Minor changes with indentation in forward.py
- Removed unused imports from tutorial notebooks
- removed a unused cell in tutorial.ipynb
- some minor changes
- added required package: ['tensorflow>=2.4.1','mne>=0.22.0', 'scipy', 'colorednoise', 'joblib', 'jupyter', 'ipykernel', 'matplotlib', 'pyvista>=0.24', 'pyvistaqt>=0.2.0', 'vtk>=9.0.1', 'tqdm']
- Removed mayavi imports in tutorials
- changed source plot functions in tutorial
- removed seaborn and mayavi as required package
- added tutorials to the repository
- changes to README
- add packages to the install requirements: colorednoise, joblib, seaborn, mayavi
- changed required python version to >=3.8.3