Checkout of HDF5 datasets

[khinsen]

Currently, aptool checkout works only on text files, not on array-like HDF5 datasets.

Question: what would be a useful format for such datasets? It needs to be easier to use (by some meaningful definition of "easy") than HDF5, because accessing the datasets via HDF5 is already possible (and straightforward at least from high-level languages).

[khinsen]

In #20, @sjdv1982 proposes .npy. That's easy to convert from and to most (but not all) HDF5 datasets, and it's easier to use in not requiring HDF5, a rather heavy dependency. It has also been proposed elsewhere in an alternative implementation of the HDF5 data model (see Experimental Directory Structure (Exdir): An Alternative to HDF5 Without Introducing a New File Format)

[sjdv1982]

Here is a quick-and-dirty tool that dumps all datasets in "data/" in .npy format.

It seems that h5py is bugged for structured arrays, but I was able to work around it.