Search code examples
pythondatasethdf5h5py

Is there a way of removing rows from a HDF5 dataset?


I have created a H5PY dataset, with around 2.1 million instances. The issue is I have filled all the rows apart from the last one. I want to remove the last row but unsure if it is feasible or safe to do.

This is a snippet of how the dataset is created:

shape = (dataset_length, args.batch_size, 2048, 1, 1)

with h5py.File(path, mode='a') as hdf5_file:
       array_40 = hdf5_file.create_dataset(
                  f'{phase}_40x_arrays',  shape, maxshape=(None, args.batch_size, 2048, 1, 1)


# either new or checkpointed file exists
# load file and create references to exisitng h5 datasets
with h5py.File(path, mode='r+') as hdf5_file:
      array_40 = hdf5_file[f'{phase}_40x_arrays']

     for i, (inputs40x, labels) in enumerate(dataloaders_dict):

          inputs40x = inputs40x.to(device)
          x40 = resnet(inputs40x)
          array_40[batch_idx, ...] = x40.cpu()

          hdf5_file.flush()

I'm not really sure if I need to copy all instances to a new dataset. I tried resizing, but that didn't work...

Cheers,


Solution

  • Here is a very simple example to demonstrate dataset.resize() for one dataset.

    import numpy as np
    import h5py
    
    arr = np.random.rand(100).reshape(20,5)
    
    with h5py.File('SO_61487687.h5', mode='a') as h5f:
         h5f.create_dataset('array1',  data=arr, maxshape=(None, 5) )
    
    with h5py.File('SO_61487687.h5', mode='r+') as h5f:
         print ('Before:', h5f['array1'].shape)
         h5f['array1'].resize(10,axis=0)
         print ('After:', h5f['array1'].shape)
         h5f.flush()