Subject noise selection and Random window pairs options

[jawardell]

This commit adds two major features to the codebase:

Option to select which subject matrix is used for noise

changes to experiment.py and usp_utils.py

1. add subject_id parameter to select and track subject noise
2. add function to select subject noise
3. generated and saved the subject noise data in a dataframe saved as pkl file (not tracked)

Option to use random window pairs for combined features

changes to experiment.py and usp_utils.py and plot.ipynb

1. added parameter to toggle using random window pairs or not
2. added functions to generate all window pairs between sampling rates
3. added function to shuffle the window pairs
4. added function to choose n windows to keep class and group distribution balanced
5. looked at class and group distribution in two shuffling strategies
   • can see this only in some commits, was deleted later on
   • was done in the plot.ipynb notebook

Additional Work

Added some (untracked) scripts to track down which subjects' jobs were dropped and compare with all subjects.
Thought about how to better read in the result files for plotting to include the subject id as a field of the dataframe.

Next steps:

• need to update slurm scripts to use window flag or not
• need to make some slurm scripts for launching many experiments with random windows
  • determine whether to use many subjects' noise or not for the random windows (probably a subset)

[jawardell]

merge window pairs branch into inter subject branch. test then merge to master.