How to read HDF5 files that have only datasets (no groups) using h5py?

Question

I have HDF5 files that I would like to open using the Python module h5py (in Python 2.7).

This is easy when I have a file with groups and datasets:

import h5py as hdf

with hdf.File(relative_path_to_file, 'r') as f:
    my_data = f['a_group']['a_dataset'].value

However, in my current situation I do not have groups. There are only datasets. Unfortunately, I cannot access my data no matter what I try. None of the following work (all break with KeyErrors or ValueErrors):

my_data = f['a_dataset'].value #KeyError

my_data = f['/a_dataset'].value #KeyError

my_data = f['/']['a_dataset'].value #KeyError

my_data = f['']['a_dataset'].value #ValueError

my_data = f['.']['a_dataset'].value #KeyError

I can remake my files to have a group if there is no solution. It really seems like there should be a solution, though...

It seems like h5py is not seeing any keys:

f.keys()
[]

Sorry, I should have included that. I get an empty list: [ ] — Joshua Zollweg, Feb 16 '15 at 18:16
Well then I doubt your HDF5 file actually contains anything. Have you tried checking with h5ls or similar? Maybe you can share the output of h5ls or show us the code that produced the HDF5 file? — gspr, Feb 16 '15 at 18:18
I can see the data when I open the file with HDHView, so I know it's there. In HDFView I can select each dataset with syntax like [/a_dataset], but that doesn't work in python h5py. — Joshua Zollweg, Feb 16 '15 at 18:28
And you're absolutely positive that you inspected `f.keys` *within* the `with hdf…` block? — gspr, Feb 16 '15 at 18:30
Yes, I got the errors when running code and I've tried subsequently in debug mode and interactively (with and without "with") — Joshua Zollweg, Feb 16 '15 at 18:31

Joshua Zollweg · Accepted Answer · 2015-02-16T20:36:53.377

4

I found the issue, which I think is an issue h5py should address.

The issue (which I originally forgot to detail in the question, now edited) is that I open the hdf5 file with a relative file path. When I use and absolute file path, everything works perfectly.

Sadly, this is going to cause me problems down the road as my code is intended to run portably on different machines...

Thanks to gspr and jimmyb for their help :-)

edited Feb 16 '15 at 20:36

answered Feb 16 '15 at 18:48

Joshua Zollweg

174
1
1
8

Its 2019, and I can confirm the issue has not been addressed yet. Thankfully for your answer... – atmaere Aug 29 '19 at 03:44

score 2 · Answer 2 · edited Mar 04 '16 at 20:32

2

It worked fine when I was using a relative path.

To write:

fileName = "data/hdf5/topo.hdf5"

with h5py.File(fileName, 'w') as f:
    dset = f.create_dataset('topography', data = z, dtype = 'float32')

To read data:

with h5py.File(fileName, 'r') as f:
    my_data = f['.']['topography'].value

edited Mar 04 '16 at 20:32

user812786

4,302
5
38
50

answered Mar 04 '16 at 18:26

Jasmine

21
4

score 1 · Answer 3 · answered Feb 16 '15 at 18:15

1

I think that this should work:

f['.']['a_dataset']

And you might try to do:

dir(f['/']) 
dir(f['.'])

answered Feb 16 '15 at 18:15

jimmyb

4,227
2
23
26

Sadly it didn't work for me. When I use f.keys() I get an empty list. – Joshua Zollweg Feb 16 '15 at 18:19
Empty lists of keys are returned for all these options. For some reason, there is no group membership... – Joshua Zollweg Feb 16 '15 at 18:26

How to read HDF5 files that have only datasets (no groups) using h5py?

3 Answers3