Currently, non-wear detection uses the GGIR algorithm, which is initially calculated over a 15-min resolution. However, we want to export non-wear flag at the same resolution as epoch1 metrics (5s). Currently, we upsample with .interpolate, causing thus it is not actually a boolean False/True value.
Options to change:
Convert to boolean
Implement new, more robust non-wear detection at finer resolution (5s, epoch1-level). This will eliminate the need for upsampling
Tasks
[ ] Task 1 - Implement the DETACH algorithm from scikit at epoch1-level resolution?
[ ] Task 2- Develop our own non-wear detection algorithm?
Description
Currently, non-wear detection uses the GGIR algorithm, which is initially calculated over a 15-min resolution. However, we want to export non-wear flag at the same resolution as epoch1 metrics (5s). Currently, we upsample with .interpolate, causing thus it is not actually a boolean False/True value.
Options to change:
Tasks
Freeform Notes
No response