Make common convenience methods for aggregating data for ``Group.members`` and ``Bundle``. about mdsynthesis HOT 2 CLOSED

This may require a thin abstraction layer between Bundle and its underlying members in order for it and Members to include the same methods.

from mdsynthesis.

Working on this now. The basic idea is that Bundle and Members will look similar in interface to a Container, having tags, categories, and data properties that allow manipulations on all members.

One point of interest: I currently have a Group.members.data.retrieve() method that will work for dataframe data by creating a multi-index concatenation of the dataframes for each member corresponding to the given handle. The first index level contains the name of the member (which is actually a minor problem, since names need not be unique (though neither must indexes)), while the second level gives the original index. This can work fine for dataframes and series, but will simply not work for Panel and Panel4D structures.

I'm not sure what reasonable aggregation scheme for these structures should be. Further, what's a reasonable structure for data that are numpy arrays or pure python (pickled) structures? Should we just through the structures in a list, essentially doing:

out = [ member.data.retrieve(handle) for member in group.members ]

Thoughts?

from mdsynthesis.