What is the cleanest way to add a field to a structured numpy array? Can it be done destructively, or is it necessary to create a new array and copy over the existing fields? Are the contents of each field stored contiguously in memory so that such copying can be done efficiently?
Add a comment
|
2 Answers
If you're using numpy 1.3, there's also numpy.lib.recfunctions.append_fields().
For many installations, you'll need to import numpy.lib.recfunctions to access this. import numpy will not allow one to see the numpy.lib.recfunctions
1 Comment
nico
Note that
append_fields creates an array of type 'numpy.ma.core.MaskedArray'. If you want an array of type 'numpy.ndarray', you can convert the result with df = np.array(df).import numpy
def add_field(a, descr):
"""Return a new array that is like "a", but has additional fields.
Arguments:
a -- a structured numpy array
descr -- a numpy type description of the new fields
The contents of "a" are copied over to the appropriate fields in
the new array, whereas the new fields are uninitialized. The
arguments are not modified.
>>> sa = numpy.array([(1, 'Foo'), (2, 'Bar')], \
dtype=[('id', int), ('name', 'S3')])
>>> sa.dtype.descr == numpy.dtype([('id', int), ('name', 'S3')])
True
>>> sb = add_field(sa, [('score', float)])
>>> sb.dtype.descr == numpy.dtype([('id', int), ('name', 'S3'), \
('score', float)])
True
>>> numpy.all(sa['id'] == sb['id'])
True
>>> numpy.all(sa['name'] == sb['name'])
True
"""
if a.dtype.fields is None:
raise ValueError, "`A' must be a structured numpy array"
b = numpy.empty(a.shape, dtype=a.dtype.descr + descr)
for name in a.dtype.names:
b[name] = a[name]
return b
1 Comment
gerrit
Can this be amended to avoid duplicating memory? (see this question)