Consider the pipeline generated from NimbusML that adds OptionalColumnTransform automatically to make sure label column is of correct type, i.e key type. When this model is loaded in ML.NET, OptionalColumnTransform is executed as TransformWrapper that is instantiated with an IDataView containing the entire pipeline until that point. Unfortunately transform wrapper applies the entire chain of transforms when it should just apply the last transformer in the chain, this holds true if we had a IDataTransform preceded by RowToRowMapperTransfom.

CC: @yaeldekel