QoL suggestion: Show trained kernels

[sqbl]

Currently, if you call opt.models[-1] (or any other member on the list of previous models), you will receive the "input" kernel-parameters. i.e. not the trained values (to get the trained kernels, we need to call opt.models[-1].kernel_ (note the underscore)). Can we somehow expose the trained parameters?