A Pandas implementation of ring buffer could be a interesting idea. That's probably out of the scope of this project but that's definitely something to consider for heterogeneous data.