2016-DLRW-brain

The Braintrust

Report in doc/reports/report.pdf and presentation in doc/presentation/main.pdf!

EEG-EMG RNN training

  • Run brain/__main__.py to configure and execute training on EMG and EEG data.
  • breze_RNN.py EMG training implementation
  • breze_EEG.py EEG training implementation
  • globals.py global static variables
  • data preprocessing: data.py (no explicit call necessary)

Configuration

Overwrite default values calling RNN_EEG/test_RNN method

  • n_neurons: number of neurons per hidden layer
  • batch_size: number of subsets
  • participant: list of experimentees to train on
  • series: list of series to train on
  • subsample: frequency to subsample EMG data
  • imp_weights_skip: number of important weights to skip
  • n_layers: number of hidden layers

Visualization

  • data_visualization.py
  • eeg_plotter.py for visualization of raw data
  • /img contains single training error plots
  • /images contains multiple trial training plots
t-SNE
  • t-SNE call from: bhtsne.py
  • tsne.py
  • FNN implementation: fnn.py
  • C implementation within /windows subdirectory

Miscellaneous

Deprecated

  • __init__.py
  • breze_RNN_old.py
  • test_getTables.py (function call test)
  • matlab_data.py (data structure analysis)
  • elman.py (alternative approach, not continued)
  • lstm.py (alternative approach, not continued)
  • rnnrbm.py (alternative approach, not continued)
  • LogReg.py (alternative approach, not continued)