save/load functions need overhaul

[dnjohnstone]

Saving/loading is currently pretty ugly - only defined for library objects and just done by some pretty basic pickle-ing. Now that diffsims is separate from pyxem it's more obvious that this needs to be addressed properly.

[pc494]

Unless somebody has very good solution that will be easy to make work I'm going to push this to 0.3.0

[dnjohnstone]

I think the solution here should be to create an hdf5 save/load interface following a pattern similar to pyxem/orix#67

Does that sound ok to others?

[hakonanes]

For this I would recommend using functions similar to hdf5group2dict() (https://github.com/pyxem/orix/blob/master/orix/io/plugins/orix_hdf5.py#L59) and dict2hdf5group() (https://github.com/pyxem/orix/blob/master/orix/io/plugins/orix_hdf5.py#L334). They have been adapted from HyperSpy's HDF5 plugin (hspy: https://github.com/hyperspy/hyperspy/blob/RELEASE_next_minor/hyperspy/io_plugins/hspy.py#L405, https://github.com/hyperspy/hyperspy/blob/RELEASE_next_minor/hyperspy/io_plugins/hspy.py#L590). By passing recursive=True to hdf5group2dict(), one can read datasets within groups within groups... into nested dictionaries. Likewise, nested dictionaries can be written to datasets within groups within groups... within an opened HDF5 group/file. This makes creating HDF5 IO quite OK.