masked arrays to deal with with missing data

[sarathnayar]

ARHMM requires continuous data and cannot have missing data. I was wondering if the repository supports the use of NumPy masked arrays to handle missing data out of the box?

[sjvenditto]

Echoing on this issue -- It would be nice for the LGSSM models (and kalman filter/smoother) to also allow for masked data. This could be implemented by including a mask variable that gets scanned in the filter/smoother that specifies valid trials, and skips trials by not updating the means/covariances (by carrying the previous value)

[ComputerBoy9000]

Lack of masking has created a ton of headaches in relation to panel data and attempting to vmap the filters- this would be a really great feature!