We have something similar with PipeOpClassWeights for "weights_learner" for classification Tasks. Do we want a PipeOp to automatically derive weights to be used in pipelines? Or do we want to refactor PipeOpClassWeights so that it can also add a measure weights column based on class? If so, would we want a separate PipeOp that does the same for regression Tasks?

Either way, this could implement different algorithms for deriving weights such as prevalence weights, inverse prevalence weights, median frequency weights, ... Look for what methods exist, are popular, and/or are implemented in other software.

We should probably think about this before adding multiclass support for PipeOpClassWeights (asked for in #798).