Question

I’m new to Python’s iterators, so maybe my language is not always correct.

I have a class wrapping a list of numpy.ndarray:

class wrapper:
    def __init__(self, myList):
        self.myList = myList

    def getArrayIterator(self):
        for arr in self.myList:
            yield arr

#set list of arrays in wrapper
myList = [rand(3,3), rand(3,3), rand(3,3)]
w = wrapper(myList)

As I understood it, the second method returns a generator.
Now I want to use that generator to loop over the list and reset the arrays to something else:

for a in w.getArrayIterator():
    a = zeros((3,4))

I was hoping to have pass by reference semantics here, but that doesn’t seem to be the case.

So I tried using Python’s send() in my getArrayIterator function:

# ...
def getArrayIterator(self):
    for arr in self.myList:
        val = (yield arr)
        if val is not None:
            arr = val
# ...

But that wont work either because:

a.send(zeros((3,4)))
  AttributeError: 'numpy.ndarray' object has no attribute 'send'

Is there a simple solution to achieve my desired behavior?
Am I missing something?

---

EDIT: It has been pointed out to me, that I should provide more Information about my actual problem. The example above is of course simplified.

I have a list of lists of numpy.ndarray representing a tensor T encapsulated in my class. When accessing an element of T: t_ijkl I need to multiply the matrices stored in the list: A(i)*B(j)*C(k)*D(l), first and last are row / col vectors.

So there is a set of A, a set of B, etc. Each of which belongs to a gridpoint in my application.

I now wanted to have an iterator over all gridpoints and iterators over the matrices associated with each gridpoint.

The first Idea that came to mind was to use iterators c++ style to read and write the matrices. But as agf pointed out below, this is not really a feasible approach. So I think I’ll use the different iterators just for read access and specialized setter methods to set new values for the matrices.

Answer 1

I’m assuming this is a simplified example, because with the class you have there seems to be no reason not to just use a normal list.

When you do

name = somelist[0]
name = 'other'

you’re pointing the name name at the object at the first index in somelist, then pointing the name name at the object 'other'. You’re not ever pointing somelist[0] at 'other'.

So in addition to send, you need to actually assign to the list:

class wrapper:
    def __init__(self, myList):
        self.myList = myList

    def getArrayIterator(self):
        for i, arr in enumerate(self.myList):
            yarr = (yield arr)
            if yarr is not None:
                self.myList[i] = yarr

#set list of arrays in wrapper
myList = [1, 2, 3, 4, 5]
w = wrapper(myList)
witer = w.getArrayIterator()
try:
    a = next(witer)
    for i in range(6, 11):
        print a,
        a = witer.send(i)
except StopIteration:
    pass
print
print w.myList