Consider a switch to custom hdf5 layout

[WardLT]

We use Pandas to write our data frames to HDF5, which records them in a way that optimizes for database-like queries. The problem is that we never use such a functionality but it costs storage space and makes the data harder to read from other languages. We should switch to a custom layout with h5py

[WardLT]

After some further reading: I'm going to stick with PyTables and use it natively via PyTables rather than via Pandas. PyTables tables are just normal H5 datasets, but with a struct/compound data type row. That's something we can read easily from other software.