Bessy Roadmapping and Backlog Grooming 15 Nov 2023

Greg’s Stream of consciousness.

Things to do priority in cleaning code:

1. Separate EEGData and ERPData - files are too big
2. Move all logging/printing to logger object. Make logger object. This will move a lot of variables around.
3. Stop array indexing in flow control, it is is hard to follow. What this means:

• Hunt & Seek thing - style fix for coding.
  
  This is an example. Want to move this away from doing this, but instead put it into a variable, then put it in the code.

4. Instead of returning tuples return objects. Tuples - get rid of these if we can, give the thing a name, and make real properties. Instead of returning tuples, can we return a single object and get names for what this is.

• E.g. give words to the data instead of just indexes.
• This is in everything…but very difficult

5. Private methods use underscores.

• Will help documentation, cause we can remove what is private which enables them to be removed from the docstrings.

6. Inject classifier at initialization. (Inject at init).
7. Start-up is slow - make this faster. Optimization in general? We are importing too much at once.

• Creating EEG data is taking a long time for initialization.

8. Unit testing - where we can.
9. Distinguish between markers and control signals in the code.
10. Backwards compatibility

• What does this look like? Who are we screwing over the moment we update python, and then fix those things.
• Might want to think about this earlier rather than later.

11. Bessy exits if EEG stream not available.

• Need to make sure we have a clean exit and close the available streams

12. Bessy crashes on missing time units member.

• There is a crash if you don’t have a TimeUnitsMember happens, then a crash occurs.
• Bessy is looking for EEG data in this case, the inlet has been created but the

13. Test out and look for ghost streams and what might exist upon crashes
14. Bessy crashes if trials ends without marked training data
15. Bessy crashes if we don’t have enough training data for predictions
16. Check if we are getting packet loss

• Handle how we deal with packet loss

17. Bessy crashes if we set iterative training to True
18. Bessy crashes if stream drops, even for a minute
19. BUS Error in Signal processing Bandpass, some NUMPY function is calling memory it isn’t allowed to.

• Unit test exists around Bandpass right now, but isn’t catching this

20. Pull out EEG as an object for dependency injection
21. Pull out Marker as an object for dependency injection

• Eli will rock it out and split commands

22. Pull out observer as an object for dependency injection
23. Provide markers asynchronously - When making predictions, need to send in a marker with a given label. We want to have an option to turn on “always on” classifier predictions out.
24. Provide prediction probabilities (if possible)