when playing with some real data (particularly with changepoint problems)
I have felt that outliers can cause problems. It would be nice to be able to
use robust regression techniques or allow non-normal residuals.
Possibly mixture of normal residuals, or allow residuals that have observation-specific
variances. This might be easier when we have a version with heterogeneous variances implemented.
when playing with some real data (particularly with changepoint problems) I have felt that outliers can cause problems. It would be nice to be able to use robust regression techniques or allow non-normal residuals.
Possibly mixture of normal residuals, or allow residuals that have observation-specific variances. This might be easier when we have a version with heterogeneous variances implemented.