Currently, our map from gains to visibilities follows a time station order.
To explain this, note that our gain parameters are represented as a flat vector of gains. The order is such that we index over the stations first and then time stamps. This makes sense if you look at each time stamp independently, as our current gain priors typically enforce. However, this order is not ideal when dealing with time-correlated gains.
For time-correlated gains, storing the vectors in a station time ordering would be nice. This means we should first place group the gains for each station and then concatenate the times together. This will simplify certain gain orderings, and I can remove the need for relative gain segmentation, which is slow and not ideal.
The roadmap
Define a gain ordering abstract type
Move the current scheme to TimeStation()
Implement the new scheme as StationTime()
Remove the ScanSeg{true}() since it is now redundant
Implement a time correlated gain prior to test this out.
Currently, our map from gains to visibilities follows a time station order. To explain this, note that our gain parameters are represented as a flat vector of gains. The order is such that we index over the stations first and then time stamps. This makes sense if you look at each time stamp independently, as our current gain priors typically enforce. However, this order is not ideal when dealing with time-correlated gains.
For time-correlated gains, storing the vectors in a station time ordering would be nice. This means we should first place group the gains for each station and then concatenate the times together. This will simplify certain gain orderings, and I can remove the need for relative gain segmentation, which is slow and not ideal.
The roadmap