Redundancy in Algorithm Types #130
Description
There is a lot of redundancy in the type hierarchy for our analytical algorithms.
Our current setup is essentially a 2D grid with option A from:
- Kalman
- Information
- Discrete
- Square Root
- Square Root Information
And option B from:
- Filter
- Smoother
- Backward predictor
- Two-filter smoother
I suggest we discard the "B" column since we can determine this by the method being called (e.g. backward_initialise).
I've been tempted in the past to remove option A and just dispatch on the state type since that is technically enough information, but I guess it's nice to keep the algorithm to specify things like how much jitter to add, whether to use the Joseph form update, what PD strategy to use etc..
The only downside of removing the "B" column is that we can no longer specify different jitter amounts for the initialisation, filtering, backward prediction etc.