Save and load HDF5 files (h5 file extension) and resolve all references recursively into suitable formats.

Complicated, nested structures like dictionaries or structured numpy arrays may be saved as they come and will be rebuild the same way on loading. Python data types like dicts, lists, tuples, ranges or slices are maintained and loaded exactly as they are saved.

python setup.py install

Usage is straight forward for loading and saving files.

To write the variables var, var2 and var3 to a new file, write

from h5deref import h5save
h5save('/file/path.h5', {'a': var, 'b': var2, 'c': var3})

Additional h5py options like compression can be passed to the datasets.

from h5deref import h5save
h5save('/file/path.h5', {'a': var}, compression='gzip')

All content from a file is loaded into a structured numpy array with

from h5deref import h5load
data = h5load('/file/path.h5')
data.var2.nested.variables  # Access easily through numpy recarray

The content can alternatively be loaded into a Python dictionary.

from h5deref import h5load
data = h5load('/file/path.h5', dict=True)
data.keys()  # dict_keys(['a', 'b', 'c'])

To speed up loading, individual keys can be specified to be loaded only.

from h5deref import h5load
data = h5load('/file/path.h5', keys=['/a', '/c/subname'])
data.dtype.names  # ('a', 'c')

These can be written into different variables directly. They are unpacked in alpha-numerical order, not in order of specification.

from h5deref import h5load
a, c = h5load('/file/path.h5', keys=['/a', '/c/subname'])[()]

MATLAB files (mat file extension) saved with v7.3 are HDF5 files under the hood. This packages understands (most of) MATLAB's data types and enables to load and write them as well, while correctly transposing the data (by identifying MATLAB files by their file extension).

These MATLAB files may thus function as convenient interface to share heavy MATLAB structs or Python dictionaries/structured numpy arrays between Python and MATLAB without any data conversion or complicated loading scripts.

from h5deref import h5save, h5load
data = h5load('/file/path.mat').data
# ...
h5save('/file/path.mat', {'data': data})

On loading, singleton dimensions are squeezed for easier indexing in Python. Saving MATLAB files will append singleton dimensions to the beginning of one-dimensional arrays for compatibility with MATLAB. When saving a previously loaded MATLAB file, it is unavoidable that one-dimensional matrices may end up transposed, i.e. (432, 1) ---loading---> (432,) ---saving---> (1, 432). For indexing in MATLAB, this does not make a difference.

Support for MATLAB data types is implemented on an as-needed basis. Loading and saving further data types may be implemented in the future.