ENH: Create 2DTransfrom.py #12723
ENH: Create 2DTransfrom.py #12723
Conversation
| arr = np.empty([2, 2]) | ||
|
|
||
| arr[0][0] = math.cos(angle) | ||
| arr[0][1] = -math.sin(angle) | ||
| arr[1][0] = math.sin(angle) | ||
| arr[1][1] = math.cos(angle) | ||
|
|
||
| return arr #This returns a rotation matrix |
There was a problem hiding this comment.
If this is going to be worth including (anywhere) in scipy, it's probably worth makign it broadcasting:
c = np.cos(angle)
s = np.sin(angle)
arr = np.empty(c.shape + (2, 2), dtype=c.dtype)
arr[..., 0, 0] = c
arr[..., 0, 1] = -s
arr[..., 1, 0] = s
arr[..., 1, 1] = cThere was a problem hiding this comment.
I have made the required changes.
|
Before making more detailed suggestions, I'm curious what @nmayorov thinks about the 2D transformations idea given all their work on the 3D/quaternion stuff in SciPy. |
Well, 2D rotations are much more simple and universally described by a single angle. Can be seen as a 3D rotations around Z (for example) axis. Providing As for the provided simple implementation -- I guess I'm about neutral on its inclusion (after implementation and docstring cleanups). There are some discussions about providing rotations in N-dimensions in a selected plane. I think it's not often used and too abstract (besides linear algebra maybe, but they definitely don't want to use scipy in their compiled code) As for rotation as part of 4D transformation matrix in homogeneous coordinates -- maybe we could go with |
Picking up from
numpy/numpy#17089
#12693
numpy/numpy#17043
I have added transformations in 2D.