Short run:
For a model without x's (no-x or dummy-x), the corresponding beta should be fixed (at 1?) i.e. not sampled.
In the long run, the existence of that beta should be concealed from the user. Currently, it needs to be in samples in order to do predictions. The beta vector could be augmented internally instead (both in computeloglik and in predict).
(set this up as a second issue when the above is complete).
Short run: For a model without x's (no-x or dummy-x), the corresponding beta should be fixed (at 1?) i.e. not sampled.
In the long run, the existence of that beta should be concealed from the user. Currently, it needs to be in samples in order to do predictions. The beta vector could be augmented internally instead (both in computeloglik and in predict). (set this up as a second issue when the above is complete).