ChangePoints kernel does not support active_dims, which means that it can only be used when the entire input is 1D, whereas it should be possible to have a changepoints kernel apply to any single dimension of the input.

ChangePoints should accept the active_dims keyword, and check that it is of length 1 only.
(Also, some error-handling in K() would be good; the tensorflow error that is produced at the moment isn't very helpful.)

Minimal example: https://stackoverflow.com/questions/59613675/gpflow-change-point-kernel-issue-with-multiple-dimensions