Random sampling #220
Random sampling #220
Conversation
| if accel == 0: | ||
| return np.arange(dim_size) # type: ignore | ||
|
|
||
| indexes = list(range(dim_size)) |
There was a problem hiding this comment.
Why pick positions over a discrete grid when uniform and gaussian pdfs are defined over continuous spaces ? We could remove dim_size entirely as it would be only relevant when pdf is an array which already contains dim_size in its shape
There was a problem hiding this comment.
That's a good point. There are two reason: 1) this function was initially written 2 years ago for creating cartesian mask, and it evolved from there 2) this enforces a stacking that lies on the cartesian grid, making it suitable for Stacked-NUFFT (2.5 D) optimization.
A solution could be to resort to the continuous case if dim_size = None , but then accel does not make a lot of sense anymore (accelerating compared to the continuous kz is not defined)
|
@Daval-G What do you think of this updated version ? also do not hesitate to check the example as well |
paquiteau
force-pushed
the
stack-random
branch
from
2ece4be to
c4dc349
Compare
paquiteau
force-pushed
the
stack-random
branch
from
c4dc349 to
630486c
Compare
|
@paquiteau I am starting again to look at this request, but I won't finish tonight. I see that a bunch of my old comments are un-answered, could you have a look ? Thx |
|
I think now I have address everything you asked for :) |
This PR adds random sampling extension for trajectories.
In particular, it adds a trajectory expansion function (
stack_random) to stack 2D trajectory following a certain distribution.([ ] Add ND sampling (e.g poisson disc))An example of stack of spiral with uniform sampling (but fully sampled in the center of k-space)
