def __init__(self, filename, groupname, index=None):
"""Initializes a h5features reader to read a group in a HDF5 file."""
# check filename
if not h5py.is_hdf5(filename):
raise IOError('{} is not a HDF5 file'.format(filename))
self.filename = filename
# open the HDF5 file for reading
self.h5file = h5py.File(self.filename, 'r')
# access to the requested group
if not groupname in self.h5file:
raise IOError('{} is not a valid group in {}'
.format(groupname, self.filename))
self.groupname = groupname
self.group = self.h5file[groupname]
# Get the version of the readed file
self.version = self._read_version()
# read the index from group if not provided
if index is None:
# Choose the good index according to file version
if self.version == '0.1':
index_class = IndexV0_1()
elif self.version == '1.0':
index_class = IndexV1_0()
else:
index_class = Index()
self.index = index_class.read(self.group)
else:
self.index = index
def write(self, data, groupname='features', append=True):
"""Write h5features data in a specified group.
Parameters
----------
- data : dict --- TODO document this!
- groupname : str, optional --- The name of the group in which
to write the data.
- append : bool, optional --- This parameter has no effect if
the *groupname* is not an existing group in the file. If
set to True (default), try to append new data in the
group. If False erase all data in the group before
writing.
"""
# shortcut from parameters
items = data['items']
times = data['times']
featu = data['features']
# Open the HDF5 file for writing/appending in the group.
with h5py.File(self.filename, mode='a') as h5file:
# Initialize an empty index
index = Index()
# The group already exists
if groupname in h5file:
group = h5file[groupname]
# want to append data, raise if we cannot
if append and not self.is_appendable_to(group, data):
raise IOError('data is not appendable to the group {} in {}'
.format(groupname, self.filename))
# want to overwrite, delete the existing group
# TODO test that
if not append:
del group
else:
# The group does not exist, create it...
group = h5file.create_group(groupname)
group.attrs['version'] = self.version
# ... and initialize it with empty datasets
featu.create_dataset(group, self.chunk_size)
items.create_dataset(group, self.chunk_size)
index.create_dataset(group, self.chunk_size)
# chunking the times depends on features chunks
times.create_dataset(group, featu.nb_per_chunk)
# writing data TODO assert no side effects here,
# e.g. writting features concat them in place...
index.write(group, data['items'], data['features'])
for dataset in ['items', 'times', 'features']:
data[dataset].write(group)
