unstructured sample data

[nsheff]

could we think of way to preserve the portability benefits of the PEP (derived attributes, etc), but switch from the tabular format to an unstructured sample definition? it would be read in peppy in the same way, making two ways to represent samples that could be united into one representation in memory. useful for data structures that don't fit well with tabular, hierarchical data.