CODEWITHSUNDEEP
pythonarraysnumpyinitialization
mcocdawc
mcocdawc

Reputation: 1867

Initialize transposed numpy array

I want to use the Singular-Value-Decomposition of matrix A.

If possible I would write:

V, S, W.T = np.linalg.svd(A)

But I can't initialise an array with its transposed. Now I have two questions:

  1. As far as I understand the python internals there is no obvious workaround for this problem. Because the call of an attribute/method of W requires the instance to be initialised. One would need something as a constructor as @property attribute.

  2. If there is no obvious workaround, which one of the following options is better/more idiomatic.

Option 1:

V, S, tmp = np.linalg.svd(A)
W = tmp.T

Option 2:

V, S, W = np.empty(...), np.empty(...), np.empty(...)
V[:, :], S[:, :], W.T[:, :] = np.linalg.svd(A)

Upvotes: 2

Views: 186

Answers (1)

user6655984
user6655984

Reputation:

Option 2 takes over 50% more time in my experiment. It's also harder to read.

Option 1 is good, but observe that W will be a view of the array tmp. This should not be a problem unless you do something that makes it one, like tmp[0,0] = 0 (which modifies W too).

I would go with

W = np.linalg.svd(A)
W = W.T

which runs in the same time as the version with tmp (and it still makes W a view) but does not create another name by which the same data can be accessed.

Upvotes: 3

Related Questions