Subclassing numarray's arrays

[Mizrandir]

I'd like to subclass numarray's array. I'd like to add some methods
and override others like __init__. I don't know how to do this and
haven't found help searching the manual or the web, can someone help?

For example imagine I just want to do something as simple as making a
subclass "NewClass" with the __init__ method overridden so that the
behaviour would be:

a = NewClass(2)
print a

[[2 2]
[2 2]]

How could that be done?

Thanks in advance, miz.

[Colin J. Williams]

Mizrandir wrote:

I'd like to subclass numarray's array. I'd like to add some methods
and override others like __init__. I don't know how to do this and
haven't found help searching the manual or the web, can someone help?

For example imagine I just want to do something as simple as making a
subclass "NewClass" with the __init__ method overridden so that the
behaviour would be:

a = NewClass(2)
print a

[[2 2]
[2 2]]

How could that be done?

Thanks in advance, miz.

Miz,

Numarray wasn't designed with subclassing in mind, but there are
workarounds for most problems.

You might try something like:
import numarray.numarraycore as _num
class NewClass(_num.NumArray):
def __init__(self, n, a):
''' n provides the length of each dimension,
a is the constant value to be plugged.
'''
arr= _num.array(sequence= n * n * [a], shape= (n, n))
self.__setstate__(arr.__getstate__())

def __repr__(self):
" Return printable representation of instance."
className= self.__class__.__name__
className= className.zfill(5).replace('0', ' ')
arr= self.copy()
arr.__class__= _num.NumArray
rep= className + _num.NumArray.__repr__(arr)[5:]
return rep

def __str__(self):
" Return a pretty printed string of the instance."
stri= self.copy()
stri.__class__= _num.NumArray
return _num.NumArray.__str__(stri)
if __name__ == '__main__':
a= NewClass(2, 2)
print a
b= NewClass(3, 4)
print `b`

I hope that this helps.

Colin W.

[Mizrandir]

First of all, thanks for helping.

Numarray wasn't designed with subclassing in mind, but there are
workarounds for most problems.
On numarray's website there is a paper by the Numarray authors saying:
"... The new approach allows arrays to be subclassed as well as ...",
so I thought it should be something easy to do.

Anyway, I hadn't read anything about numarraycore, is it explained
anywhere. For example, I don't understand what the statement
"self.__setstate__(arr.__getstate__())" does.

You might try something like:
import numarray.numarraycore as _num
class NewClass(_num.NumArray):
def __init__(self, n, a):
''' n provides the length of each dimension,
a is the constant value to be plugged.
'''
arr= _num.array(sequence= n * n * [a], shape= (n, n))
self.__setstate__(arr.__getstate__())

def __repr__(self):
" Return printable representation of instance."
className= self.__class__.__name__
className= className.zfill(5).replace('0', ' ')
arr= self.copy()
arr.__class__= _num.NumArray
rep= className + _num.NumArray.__repr__(arr)[5:]
return rep

def __str__(self):
" Return a pretty printed string of the instance."
stri= self.copy()
stri.__class__= _num.NumArray
return _num.NumArray.__str__(stri)
if __name__ == '__main__':
a= NewClass(2, 2)
print a
b= NewClass(3, 4)
print `b`

miz.

[Colin J. Williams]

Mizrandir wrote:

First of all, thanks for helping.

Numarray wasn't designed with subclassing in mind, but there are
workarounds for most problems.

On numarray's website there is a paper by the Numarray authors saying:
"... The new approach allows arrays to be subclassed as well as ...",
so I thought it should be something easy to do.

I don't remember that, but it is not as dificult as it looks. Just try
out the code below, nothing could be simpler than the __init__ method
below.

The __str__ and __repr__ are just tools which can be used with any subclass.
Anyway, I hadn't read anything about numarraycore, is it explained
anywhere. For example, I don't understand what the statement
"self.__setstate__(arr.__getstate__())" does.
NumArray creates a number of attributes to provide information about the
shape of an array, the data type of its elements and to point to the
memory buffer containing the binary data etc. The expression above just
transfers the values from arr (which is an instance of NumArray) to
self (which is an instance of NewClass).

You might try something like:
import numarray.numarraycore as _num
class NewClass(_num.NumArray):
def __init__(self, n, a):
''' n provides the length of each dimension,
a is the constant value to be plugged.
'''
arr= _num.array(sequence= n * n * [a], shape= (n, n))
self.__setstate__(arr.__getstate__())

def __repr__(self):
" Return printable representation of instance."
className= self.__class__.__name__
className= className.zfill(5).replace('0', ' ')
arr= self.copy()
arr.__class__= _num.NumArray
rep= className + _num.NumArray.__repr__(arr)[5:]
return rep

def __str__(self):
" Return a pretty printed string of the instance."
stri= self.copy()
stri.__class__= _num.NumArray
return _num.NumArray.__str__(stri)
if __name__ == '__main__':
a= NewClass(2, 2)
print a
b= NewClass(3, 4)
print `b`

miz.

Please let me know if I can provide any further information. numarray
is a good tool, but there is a small hurdle if one wishes to subclass.

Colin W.