无法访问返回的 h5py object 实例

Question

I have a very weird issue here. I have 2 functions: one which reads an HDF5 file created using h5py and one which creates a new HDF5 file which concatenates the content returned by the former function.

def read_file(filename):
    with h5py.File(filename+".hdf5",'r') as hf:

        group1 = hf.get('group1')
        group1 = hf.get('group2')            
        dataset1 = hf.get('dataset1')
        dataset2 = hf.get('dataset2')
        print group1.attrs['w'] # Works here

        return dataset1, dataset2, group1, group1

And the create file function

def create_chunk(start_index, end_index):

    for i in range(start_index, end_index):
        if i == start_index:
            mergedhf = h5py.File("output.hdf5",'w')
            mergedhf.create_dataset("dataset1",dtype='float64')
            mergedhf.create_dataset("dataset2",dtype='float64')

            g1 = mergedhf.create_group('group1')
            g2 = mergedhf.create_group('group2')

    rd1,rd2,rg1,rg2 = read_file(filename)

    print rg1.attrs['w'] #gives me <Closed HDF5 group> message

    g1.attrs['w'] = "content"
    g1.attrs['x'] = "content"
    g2.attrs['y'] = "content"
    g2.attrs['z'] = "content"
    print g1.attrs['w'] # Works Here
return mergedhf.get('dataset1'), mergedhf.get('dataset2'), g1, g2

def calling_function():
    wd1, wd2, wg1, wg2 = create_chunk(start_index, end_index)
    print wg1.attrs['w'] #Works here as well

Now the problem is, the dataset and the properties from the new file created and represented by wd1, wd2, wg1 and wg2 can be accessed by me and I can access the attribute data but i cant do the same for which I have read and returned the value for.

Can anyone help me fetch the values of the dataset and group when I have returned the reference to the calling function?

Answer 1

The problem is in read_file , this line:

with h5py.File(filename+".hdf5",'r') as hf:

This closes hf at the end of the with block, ie when read_file returns. When this happens, the datasets and groups also get closed and you can no longer access them.

There are (at least) two ways to fix this. Firstly, you can open the file like you do in create_chunk :

hf = h5py.File(filename+".hdf5", 'r')

and keep the reference to hf around as long as you need it, before closing it:

hf.close()

The other way is to copy the data from the datasets in read_file and return those instead:

dataset1 = hf.get('dataset1')[:]
dataset2 = hf.get('dataset2')[:]

Note that you can't do this with the groups. The file needs to be open for as long as you need to do things with the groups.

Answer 2

Adding to @Yossarian's answer

The problem is in read_file, this line: with h5py.File(filename+".hdf5",'r') as hf: This closes hf at the end of the with block, ie when read_file returns. When this happens, the datasets and groups also get closed and you can no longer access them.

For those who come across this and are reading a scalar dataset make sure to index using [()] :

scalar_dataset1 = hf['scalar_dataset1'][()]

---

Preface

I had a similar issue as OP resulting in a return value of <closed hdf5 dataset> . However, I would get a ValueError when attempting to slice my scalar dataset with [:] .

"ValueError: Illegal slicing argument for scalar dataspace"

Indexing with [()] along with @Yossarian's answer helped solve my problem.