Using a slice to label a dimension with HDF5

regref_dim_scales.py

import os
import h5py
import numpy as np

# Create compound data type
# a table with columns channel ID, xyz-coordinates, and impedence
coord_dt = np.dtype([('x', np.float), ('y', np.float), ('z', np.float)])
ch_dt = np.dtype([("id", np.int), ("coord", coord_dt), ("imp", np.float)])

# Make a fake channel table
a = np.zeros((10,), dtype=ch_dt)
for i in range(len(a)):
    a[i]['id'] = i
    a[i]['coord'][0] = 1.0*i
    a[i]['coord'][1] = 2.0*i
    a[i]['coord'][2] = 3.0*i
    a[i]['imp'] = 0.1*i

# Now use these things to do fun stuff!
fname ='test_cpd_dtype.h5'
if os.path.exists(fname):
    os.remove(fname)
f = h5py.File(fname)

# Cache the type to file in case we want to use it throughout the file
f['channel_row_t'] = ch_dt
# Get the named type from file so we can use it
ch_row_dt = f['channel_row_t']

# Write our fake channel table to disk using
# our named (aka 'committed') compound type
channels = f.create_dataset('channels', (10,), dtype=ch_row_dt)
channels[:] = a

# Create some fake voltage recordings
np.random.seed(1)
voltage = f.create_dataset('voltage', data=np.random.random((3,20)), dtype=np.float)

# Create dimensions scales from a region reference
## First create a special data type for a region reference dataset
regref_dt = h5py.special_dtype(ref=h5py.RegionReference)
## Next create the dataset...
voltage_scale = f.create_dataset('voltage_channels', shape=(), dtype=regref_dt)
## and fill it in with a region reference
vchannels = [3,7,9]
voltage_scale[...] = channels.regionref[vchannels]
## Now turn label this thing as a dimension scale...
voltage.dims.create_scale(voltage_scale, 'Channels')
## and then add it to our voltage data
voltage.dims[0].attach_scale(voltage_scale)
f.close()

# Verify that we can read this back in and identify what we did
f = h5py.File(fname, 'r')
voltage = f['voltage']
print("'voltage' dataset should have channels from %s" % ", ".join(str(x) for x in vchannels))
vchannels = voltage.dims[0]['Channels'][()]
if isinstance(vchannels, h5py.RegionReference):
    print('dimension scale is region reference, retrieving data.')
    vchannels = f[vchannels][vchannels]
print("'voltage' dataset has data from channels %s" % ", ".join(str(x) for x in vchannels['id']))
f.close()