1. 19 Sep, 2018 3 commits
  2. 12 Sep, 2018 2 commits
  3. 10 Sep, 2018 2 commits
  4. 19 Jan, 2018 2 commits
  5. 07 Dec, 2017 3 commits
  6. 05 Dec, 2017 1 commit
  7. 20 Nov, 2017 3 commits
  8. 17 Nov, 2017 2 commits
    • Jussi Lindgren's avatar
      f5c63f8b
    • Jussi Lindgren's avatar
      iscellaneous refactoring and fixes · 1bf66016
      Jussi Lindgren authored
      - Added some basic head/generator/pipeline initializations into config/
      - Refactored the scripts to use these initializers when possible instead of
      copy-pasting the code
      - Cleaned up proc_cincotti() to use proc_inverse_transform() that also
      does the ROI selection
      - Made 'classifytest.m' cleaner to be more illustrative of the flow
      - core_bci_simulator summary now includes mention of looped-over parameters
      for pipelines as well, not only for generators.
      - where-functions now return also a layer value telling what the indexes
      refer to (volume or surface)
      - Data generator will now use image plotting in the volume if there are too
      many sources to visualize
      - Fixed visu in core_timeline
      - Refactored extract_features_inverse into proc_inverse_transform.
      - Changed proc_inverse_transform parameters to be more consistent with conventions
      - Added mean and var spec to the gaussian noise generator
      - Fixed dataset to EEGLAB conversion
      - Fixed heuristic ROI for nonconstrained models
      - Removed standalone summarize_results as its member funct of bci_simulator.
      1bf66016
  9. 10 Nov, 2017 3 commits
    • Jussi Lindgren's avatar
      Refactored a more clear split of when and what · 8653928a
      Jussi Lindgren authored
      There is now a clear dichotomy that events are created first,
      and then the effect generators react to the events on the timeline.
      
      - Added event generators: when_always, when_random, when_trials
      - Signal generators are provided a binary mask which tells them when
      they are supposed to be active. As some masks can be impulse masks
      (zero-duration events), its up to the generator to decide how to
      react to the mask.
      - Event times are now specified in terms of sample index,
      this has the effect that timeline must know the sampling rate
      - Class labels are no longer coded in the dataset as given
      by the generator. Instead the user tells the pipeline which
      events mark the classes
      - Refactored csp implementation functions to the base class
      - Split BCI Comp IV generators to their semantic components,
      cleaner code, easier to understand
      - Updated the example scripts to work with the new conventions
      - classifytest.m now illustrates the more usual 2 class Motor Imagery
      - struct_to_list() now accepts empty structs
      - Misc small changes in option names etc
      8653928a
    • Jussi Lindgren's avatar
      Moved some contributed files to old/ · b9181da8
      Jussi Lindgren authored
      b9181da8
    • Jussi Lindgren's avatar
      Moved currently nonworking code to old/ · 344174a0
      Jussi Lindgren authored
      - in the future the code should be ported to current conventions.
      344174a0
  10. 18 Jul, 2017 3 commits
  11. 11 Jul, 2017 1 commit
  12. 29 Mar, 2017 1 commit
    • Jussi Lindgren's avatar
      Added disk cache to the BCI Simulator · 34a16022
      Jussi Lindgren authored
      - Using the cache requires jvm at the moment
      - When the cache is enabled, enlarging an experiment
        is simple: Just enlarge the set of parameters you're
        interested in and rerun the experiment. Number of
        iterations can be increased as well.
      - Fixed the data generator not keeping the volume data
        even if requested
      - Random seed is now shuffled on experiment startup by default
      - Saving of intermediate results now only happens if more
        than 10 minutes of computation passed since the last save.
      - Intermediate results are now saved under the cache/ folder
      - Fixed issue in the result merge routine
      34a16022
  13. 28 Mar, 2017 3 commits
  14. 27 Mar, 2017 2 commits
  15. 24 Mar, 2017 1 commit
  16. 22 Mar, 2017 7 commits
  17. 10 Mar, 2017 1 commit
    • Jussi Lindgren's avatar
      Various object oriented refactorings · a3619235
      Jussi Lindgren authored
      - All core classes now have the same kind of simple usage logic:
        construct object first with desired parameters, then call
        appropriate member functions.
      - All core classes can be asked for their actual parameters after
        the construction using member function get_params().
      - Refactored run_experiment() into class_bci_simulator.m
      - Params controlling simulator are now given as key,value pairs
        like most parameters in the platform.
      - Refactored print_summary.m to class_bci_simulator.m
      - Tinkered in-code documentation a bit
      a3619235