asarray() for power spaces

[adler-j]

This could be implemented as calling asarray() on all subspaces, creating a N+1:d array. Another option is to create a linear array.

[kohr-h]

Or both, using an optional boolean flag contiguous or linear_arr to differentiate.

[adler-j]

That is definitely a decent middle way.

[kohr-h]

Depends a bit on what one expects when asking a power space vector to become an array. Syntactically, one could index the vector with a single integer, vec[0], which then gave the component vector. The corresponding array would be the n-dim. one so you can use the same syntax arr[0] there to get the same as vec[0].asarray(). Hence, by default, that makes most sense IMO. An option (maybe called ravel or flat) to flatten the array would be good to have for certain applications, I guess.

[adler-j]

I agree, but we should also note that the ndarray version only works for "PowerSpace" type spaces, what about unmatched pairs? ProductSpace(Rn(3), Rn(5))?

[kohr-h]

Correct, so then it's clear that only the flat version will fly. Any reason to make an optional n-dim array version for power spaces? If not, we don't differentiate.

[adler-j]

Users could anyway themselves:

>>> x.asarray().reshape(...)

If they want nd version.

Another argument for the linear version is that we can make it recursive, so it will work on productspaces of productspaces. One interesting question is how we handle unmatched dtypes in this case, I suppose using numpy.find_common_type would make the most sense.

[kohr-h]

That's an interesting issue. Using the find_common_type sounds like a good idea. Tests will reveal if that's true.