Closed mcusi closed 1 year ago
Not currently, but we have discussed it in https://github.com/probml/dynamax/issues/99. In theory it's pretty straightforward to allow missing data -- we just make the corresponding emission potentials zero. The practical challenge was how to take gradients wrt model parameters with missing data (see issue above for more detail). I'll close this as a duplicate, but note that this is the second time in 2 weeks that someone has asked about missing data support, so we should make it happen!
Is there a way to indicate that some observations in a sequence are missing? Thanks.