$importance() for graphlearner

[mb706]

when prepending a learner with some preprocessing we still might need to access the learner's $importance

[mb706]

This is easy when only one learner is encapsulated, but what if

• Multiple learners are encapsulated
• Something else that credibly gives some importance measure is encapsulated, e.g. a filter

Possible solution: the GraphLearner gets some extra hyperparameter "which.importance" with levels = all PipeOps that have an $importance slot. The PipeOpLearner(CV) would need to have an importance slot whenever what the encapsulate has one (how can this be done during construction?), and other PipeOps can also have $importance slots in principle.

[HongxiangXu]

I don't know whether you have fixed it, as I still face the same problem. Any suggestion?