No that we implement a lot of different stuff we should take a moment and see what is really essential for someone else to use this package and remove the rest in the published version. I see the following functionality:
- Uni-/Multivariate (and hence high-dimensional) and regression data should be handled
- User defined loss functions for all kinds of data possible (with possible updates)
- Glasso (package) based loss functions already implemented
- Cross-Validation over delta and one loss function specific param (e.g. Lambda)
- Section search (with variance) and line search
- Handling of missing values
- optional: Separate covariance estimation, as proposed by Solt?
- optional: Some more diagnostics / graphics capabilities -> visualize changepoints somehow?
Anything to add?
No that we implement a lot of different stuff we should take a moment and see what is really essential for someone else to use this package and remove the rest in the published version. I see the following functionality:
Anything to add?