def test_descriptor_setitem(self):
dd = HDF5DataDescriptor(self.hdf5_file, '/g/a2')
self.assertEqual(dd.dshape, datashape.dshape('2, 3, int64'))
dd[1, 2] = 10
self.assertEqual(dd_as_py(dd[1, 2]), 10)
dd[1] = [10, 11, 12]
self.assertEqual(dd_as_py(dd[1]), [10, 11, 12])
def test_descriptor_getitem_types(self):
dd = HDF5DataDescriptor(self.hdf5_file, '/g/a2')
self.assertEqual(dd.dshape, datashape.dshape('2, 3, int64'))
# Indexing should produce DyNDDataDescriptor instances
self.assertTrue(isinstance(dd[0], DyNDDataDescriptor))
self.assertEqual(dd_as_py(dd[0]), [1, 2, 3])
self.assertTrue(isinstance(dd[1, 2], DyNDDataDescriptor))
self.assertEqual(dd_as_py(dd[1, 2]), 1)
def test_descriptor_append(self):
dd = HDF5DataDescriptor(self.hdf5_file, '/t1')
tshape = '2, { f0 : int32; f1 : int64; f2 : float64 }'
self.assertEqual(dd.dshape, datashape.dshape(tshape))
dd.append([(10, 11, 12)])
dvals = {'f0': 10, 'f1': 11, 'f2': 12.}
rvals = dd_as_py(dd[2])
is_equal = [(rvals[k] == dvals[k]) for k in dvals]
self.assertEqual(is_equal, [True] * 3)
def test_descriptor_iter_types(self):
dd = HDF5DataDescriptor(self.hdf5_file, '/a1')
self.assertEqual(dd.dshape, datashape.dshape('2, 3, int32'))
# Iteration should produce DyNDDataDescriptor instances
vals = []
for el in dd:
self.assertTrue(isinstance(el, DyNDDataDescriptor))
self.assertTrue(isinstance(el, IDataDescriptor))
vals.append(dd_as_py(el))
self.assertEqual(vals, [[1, 2, 3], [4, 5, 6]])
def load_blaze_subcarray(conf, cdir, subcarray):
import tables as tb
from blaze.datadescriptor import HDF5DataDescriptor
with tb.open_file(cdir.fname, 'r') as f:
try:
dparr = f.get_node(f.root, subcarray, 'Leaf')
except tb.NoSuchNodeError:
raise RuntimeError(
'HDF5 file does not have a dataset in %r' % dp)
dd = HDF5DataDescriptor(cdir.fname, subcarray)
return blaze.array(dd)
def test_basic_object_type(self):
self.assertTrue(issubclass(HDF5DataDescriptor, IDataDescriptor))
dd = HDF5DataDescriptor(self.hdf5_file, '/a1')
# Make sure the right type is returned
self.assertTrue(isinstance(dd, IDataDescriptor))
self.assertEqual(dd_as_py(dd), [[1, 2, 3], [4, 5, 6]])
def load_blaze_array(conf, dir):
"""Loads a blaze array from the catalog configuration and catalog path"""
# This is a temporary hack, need to transition to using the
# deferred data descriptors for various formats.
fsdir = conf.get_fsdir(dir)
if not path.isfile(fsdir + '.array'):
raise RuntimeError('Could not find blaze array description file %r' %
(fsdir + '.array'))
with open(fsdir + '.array') as f:
arrmeta = yaml.load(f)
tp = arrmeta['type']
imp = arrmeta['import']
ds_str = arrmeta.get('datashape') # optional. HDF5 does not need that.
if tp == 'csv':
with open(fsdir + '.csv', 'r') as f:
rd = csv.reader(f)
if imp.get('headers', False):
# Skip the header line
next(rd)
dat = list(rd)
arr = nd.array(dat, ndt.type(ds_str))[:]
return blaze.array(arr)
elif tp == 'json':
arr = nd.parse_json(ds_str, nd.memmap(fsdir + '.json'))
return blaze.array(arr)
elif tp == 'hdf5':
import tables as tb
from blaze.datadescriptor import HDF5DataDescriptor
fname = fsdir + '.h5' # XXX .h5 assumed for HDF5
with tb.open_file(fname, 'r') as f:
dp = imp.get('datapath') # specifies a path in HDF5
try:
dparr = f.get_node(f.root, dp, 'Leaf')
except tb.NoSuchNodeError:
raise RuntimeError('HDF5 file does not have a dataset in %r' %
dp)
dd = HDF5DataDescriptor(fname, dp)
return blaze.array(dd)
elif tp == 'npy':
import numpy as np
use_memmap = imp.get('memmap', False)
if use_memmap:
arr = np.load(fsdir + '.npy', 'r')
else:
arr = np.load(fsdir + '.npy')
arr = nd.array(arr)
arr = blaze.array(arr)
ds = datashape.dshape(ds_str)
if not compatible_array_dshape(arr, ds):
raise RuntimeError(
('NPY file for blaze catalog path %r ' +
'has the wrong datashape (%r instead of ' + '%r)') %
(arr.dshape, ds))
return arr
elif tp == 'py':
ds = datashape.dshape(ds_str)
# The script is run with the following globals,
# and should put the loaded array in a global
# called 'result'.
gbl = {
'catconf': conf, # Catalog configuration object
'impdata': imp, # Import data from the .array file
'catpath': dir, # Catalog path
'fspath': fsdir, # Equivalent filesystem path
'dshape': ds # Datashape the result should have
}
if py2help.PY2:
execfile(fsdir + '.py', gbl, gbl)
else:
with open(fsdir + '.py') as f:
code = compile(f.read(), fsdir + '.py', 'exec')
exec(code, gbl, gbl)
arr = gbl.get('result', None)
if arr is None:
raise RuntimeError(
('Script for blaze catalog path %r did not ' +
'return anything in "result" variable') % (dir))
elif not isinstance(arr, blaze.Array):
raise RuntimeError(
('Script for blaze catalog path %r returned ' +
'wrong type of object (%r instead of ' + 'blaze.Array)') %
(type(arr)))
if not compatible_array_dshape(arr, ds):
raise RuntimeError(
('Script for blaze catalog path %r returned ' +
'array with wrong datashape (%r instead of ' + '%r)') %
(arr.dshape, ds))
return arr
else:
raise ValueError(
('Unsupported array type %r from ' + 'blaze catalog entry %r') %
(tp, dir))