-
Notifications
You must be signed in to change notification settings - Fork 15
Description
This should be more a dicussion than an issue, and it has been opened only because the discussion zone is not activated here.
Out of my proper experience in the need of LISA data containers, I would love to advocate amending the DataResidualArray class in the following manners,
-
Separate the containers for timeseries and frequency series. Each of them would be a subclass of a general base data container class to inherit common methods. A natural consequence of this would be, the transformation from timeseries to frequency series (and vice versa) would no longer be implicit, but rather explicitly performed when required. This has the advantage of,
- allowing to easily apply arbitrary tapering window
- allowing easy extension to time-frequency representations
- conceptually decoupling measurements from theories (waveform model)
-
Provide interfaces that aggregate signals of multiple channels and harmonics. It would be easier for the likelihood computation classes to work with these interfaces since the information on channels (and in some scenarios, the harmonics as well) is needed by the likelihood calculation. Note the 'template' aggregation model could be different from that for 'data', as the latter models the measurements so that they are recorded at the same time grid and can be represented on the same frequency (or time-frequency) grid, whereas the former can be generated on grids of arbitrary choice, possibly different across harmonics according to the need.