Databases Reference
In-Depth Information
>>>
ref_out
=
dset
.
regionref
[
10
:
90
]
>>>
ref_out
<HDF5 region reference>
Like object references, region references are generally opaque. The only useful aspects
are the shape of the dataspace (the same as the parent dataset), and the shape of your
selection:
>>>
dset
.
regionref
.
shape
(
ref_out
)(
100
,)
>>>
dset
.
regionref
.
selection
(
ref_out
)(
80
,)
This represents the shape of your selection; in other words, if you had applied your
slicing arguments directly to the dataset, it's the shape of the array that would have been
returned from HDF5.
Once you've got a region reference, you can use it directly as a slicing argument to
retrieve data from the dataset:
>>>
data
=
dset
[
ref_out
]
>>>
data
array([ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0.], dtype=float32)
>>>
data
.
shape
(80,)
Fancy Indexing
Keep in mind that if you're using “fancy” indexing methods (like Boolean arrays), then
the
shape
will
always
be one-dimensional. This mimics the behavior of NumPy for such
selections.
For example, suppose we had a little two-dimensional array, which we populated with
some random numbers:
>>>
dset_random
=
f
.
create_dataset
(
'small_example'
,
(
3
,
3
))
>>>
dset_random
[
...
]
=
np
.
random
.
random
((
3
,
3
))
>>>
dset_random
[
...
]
array([[ 0.32391435, 0.070962 , 0.57038087],
[ 0.1530778 , 0.22476801, 0.7758832 ],
[ 0.75768745, 0.73156554, 0.3228527 ]], dtype=float32)
We could create a Boolean array representing the entries greater than 0.5:
>>>
index_arr
=
dset_random
[
...
]
>
0.5
>>>
index_arr
array([[False, False, True],
