Geek Answers Handbook

Subclassification of numpy scalar types

I'm trying to subclass numpy.complex64 to use the way numpy stores data (the adjacent, alternating real and imaginary parts), but uses my own routines __add__ , __sub__ , ....

My problem is that when I create numpy.ndarray by setting dtype=mysubclass , I get a numpy.ndarray with dtype='numpy.complex64' instead, which leads to the fact that numpy does not use my own functions for additions, subtractions, etc.

Example:

 import numpy as np class mysubclass(np.complex64): pass a = mysubclass(1+1j) A = np.empty(2, dtype=mysubclass) print type(a) print repr(A)

Conclusion:

 <class '__main__.mysubclass'> array([ -2.07782988e-20 +4.58546896e-41j, -2.07782988e-20 +4.58546896e-41j], dtype=complex64)'

Does anyone know how to do this?

Thanks in advance - Soren

+6
source share
2 answers

A system like NumPy is only intended to be extended from C via the PyArray_RegisterDataType function. It may be possible to access this function from Python using ctypes, but I would not recommend it; it’s better to write an extension in C or Cython, or a subclass of ndarray , as @seberg describes.

Here is a dtype example in the NumPy source tree: newdtype_example / floatint.c . If you are in Pyrex, reference.pyx in the pytables source might be worth a look.

+3
source

Note that scalars and arrays are very different from numpy. np.complex64 (this is a 32-bit float, just to be noted, not double precision). You cannot change such an array, you will need to subclass the array and then override it __add__ and __sub__ .

If that’s all you want to do, it should just work in reverse order http://docs.scipy.org/doc/numpy/user/basics.subclassing.html , since subclassing the array is not so simple.

However, if you want to use this type also as a scalar. For example, you want to index scalars, this is getting harder, at least for the time being. You can get a little more by specifying __array_wrap__ to convert scalars to your own scalar type for some reduction functions, so that indexing works in all cases, it seems to me that you can define your own __getitem__ at the moment.

In all cases, with this approach, you are still using a complex data type, and all functions that are not explicitly overridden will still behave the same. @ecatmur mentioned that you can create new data types from the C side if this is really what you want.

+3
source

More articles:


All Articles