HDF5中的XML文件，h5py

Question

I am using h5py to save data (float numbers), in groups. In addition to the data itself, I need to include an additional file (an .xml file, containing necessary information) within the hdf5. How do i do this? Is my approach wrong?

f = h5py.File('filename.h5')
f.create_dataset('/data/1',numpy_array_1)
f.create_dataset('/data/2',numpy_array_2)
.
.

my h5 tree should look thus:

/ 
/data
/data/1 (numpy_array_1)
/data/2 (numpy_array_2)
.
.
/morphology.xml (?)

Answer 1

One option is to add it as a variable-length string dataset.

http://code.google.com/p/h5py/wiki/HowTo#Variable-length_strings

Eg:

import h5py
xmldata = """<xml>
<something>
    <else>Text</else>
</something>
</xml>
"""

# Write the xml file...
f = h5py.File('test.hdf5', 'w')
str_type = h5py.new_vlen(str)
ds = f.create_dataset('something.xml', shape=(1,), dtype=str_type)
ds[:] = xmldata
f.close()

# Read the xml file back...
f = h5py.File('test.hdf5', 'r')
print f['something.xml'][0]

Answer 2

If you just need to attach the XML file to the hdf5 file, you can add it as an attribute to the hdf5 file.

xmlfh = open('morphology.xml', 'rb')
h5f.attrs['xml'] = xmlfh.read()

You can access the xml file then like this:

h5f.attrs['xml']

Notice, also, that you can't store attributes larger than 64K, you may want to compress the file before attaching. You can have a look at compressing libraries in the standard library of Python.

However, this doesn't make the information in the XML file very accessible. If you want to associate the metadata of each dataset to some metadata in the XML file, you could map it as you need using an XML library like lxml . You can also add each field of the XML data as a separate attribute so that you can query datasets by XML field, this all depends on what you have in the XML file. Try to think about how you would like to retrieve the data later.

You may also want to create groups for each xml file with its datasets and put it all in a single hdf5 file. I don't know how large are the files you are managing, YMMV.