Added
- All dataset class have a new property called `data_ss` referring to the single sensor that should be used in the
algorithms.
This was added to move all data related config (i.e. which sensor to use in a pipeline) to the dataset class, making
it easier to implement dataset agnostic pipelines (https://github.com/mobilise-d/mobgap/pull/119)
- A evaluation pipeline for GSD (https://github.com/mobilise-d/mobgap/pull/124)
- ML based LR classification (https://github.com/mobilise-d/mobgap/pull/106)
- A evaluation/optimization pipeline for LRC (https://github.com/mobilise-d/mobgap/pull/106)
Changed
- The loaded reference data now has stricter dtypes (https://github.com/mobilise-d/mobgap/pull/119)
- Renamed LRD (left-right-detection) to LRC (left-right-classification) (https://github.com/mobilise-d/mobgap/pull/106)
- For GSD and IC evaluation metrics it is now possible to configure what happens in case of 0-division
(https://github.com/mobilise-d/mobgap/pull/127)
Fixed
- Loading the reference data from a trial without identified WBs will not raise an error anymore, but will correctly
return an empty DataFrame (https://github.com/mobilise-d/mobgap/pull/119)
- We use operating system independent pandas dtypes everywhere (https://github.com/mobilise-d/mobgap/pull/118)
Development
- RTD previews are now build for PRs.