ArrayCollection is a NumPy ndarray-like type which stores an ordered set of
named NumPy arrays with a common shape but which may be different dtypes. It
supports NumPy-style indexing of the arrays as a group.
>>> a = ArrayCollection({'age': [8, 10, 7, 8], 'weight': [10, 20, 13, 15]})
>>> a
ArrayCollection([( 8, 10), (10, 20), ( 7, 13), ( 8, 15)],
dtype=[('age', '<i8'), ('weight', '<i8')])
>>> a[:2]
ArrayCollection([( 8, 10), (10, 20)],
dtype=[('age', '<i8'), ('weight', '<i8')])
>>> a['age']
array([ 8, 10, 7, 8])An ArrayCollection has a very similar interface as a NumPy structured
ndarray and they can often be used interchangeably. But while NumPy's
structured arrays are designed for lowlevel manipulation of binary blobs and
for interfacing with C programs, ArrayCollection is meant for simple
manipulation of multidimensional, multi-datatype datasets. At the technical
level, ArrayCollection is a container for a set of ndarrays, while structured
ndarrays are arrays of C-structs. This means that users of ArrayCollection
do not need to worry about 'padding' bytes present in structured arrays, and
ArrayCollection will often be more cache-friendly because the individual arrays are not strided in memory as the fields are are in structured ndarrays.
ArrayCollection can be used as a "lite" version of Pandas DataFrame or
Xarray Dataset. You are encouraged to consider using these other projects
instead as they provide many additional high-level features useful for tabular
data analysis, such as named axes. ArrayCollection in contrast provides a simpler more direct interface to a set of ndarrays.
ArrayCollection uses the same datatype specification as NumPy's structured arrays.
One can use np.broadcast_to to broadcast the input arrays to the
same shape in a memory-saving way, though this makes the arrays readonly.
See the np.broadcast_to docstring for details.