Cython: memory views on `numpy` arrays lose `numpy` array features?

Question

Consider the following example:

cdef test_function():
    cdef:
        double[:] p1 = np.array([3.2, 2.1])
        double[:] p2 = np.array([0.9, 6.])

    return p1-p2

If used, it returns the following error:

Error compiling Cython file:
------------------------------------------------------------
...
cdef test_function():
    cdef:
        double[:] p1 = np.array([3.2, 2.1])
        double[:] p2 = np.array([0.9, 6.])

    return p1-p2
            ^
------------------------------------------------------------

cython_cell_v3.pyx:354:13: Invalid operand types for '-' (double[:]; double[:])

If I am using numpy arrays to initialize the memory view, how can I go about using its functionality? Do I have to somehow do some dereferencing on the memoryviews?

hpaulj · Accepted Answer · 2015-03-10 05:43:00Z

2

This works:

cpdef test_function():
    cdef:
        double[:] p1 = np.array([3.2, 2.1])
        double[:] p2 = np.array([0.9, 6.])

    # return p1-p2
    cdef int I
    I = p1.shape[0]
    for i in range(I):
        p1[i] -= p2[i]
    return np.asarray(p1)
print "Test _function", test_function()

I iterate on the arrays as though they were 'c' arrays. And without the final np.asarray, it will just display

>>> memview.test_function()
<MemoryView of 'ndarray' at 0xb60e772c>

See also the example in http://docs.cython.org/src/userguide/memoryviews.html#comparison-to-the-old-buffer-support

I tried a different function:

cpdef test_function1(x):
    cdef:
        int i, N = x.shape[0]
        double[:] p1 = x
    for i in range(N):
        p1[i] *= p1[i]
    return np.asarray(p1)*2

x = np.arange(10.)
print "test_function1 return", test_function1(x)
print "x after test_function1", x

As expected, after the function x is x**2. But what the function returns is 2*x**2.

I modify p1 directly, but end up modifying x as well. I think of p1 as a view of x, but one with reduced functionality. np.asarray(p1) gives it a numpy functionality, so I can perform an array * on it and return the result (without further modifying x).

If instead I'd finished the function with:

out = np.asarray(p1)
out *= 2
return out

I end up modifying the original x as well. out is a numpy view on x. out behaves like an array because it is one, not because of some distant link to x.

edited Mar 10, 2015 at 5:43

answered Mar 9, 2015 at 5:25

hpaulj

233k14 gold badges260 silver badges392 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

hpaulj Over a year ago

yes; I'm correcting it. I was paying more attention as to whether I needed to create a return array or memory view than to arithmetic operator.

bzm3r Over a year ago

I don't know if I can accept this as an answer simply because I wasn't looking for any solution, but in particular I was wondering how typed memoryviews of numpy arrays could still remember that they were numpy arrays?

hpaulj Over a year ago

I added a function that shows the memoryview is a view with reduced functionality.

Collectives™ on Stack Overflow

Cython: memory views on `numpy` arrays lose `numpy` array features?

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related