Stream: helpdesk (published)

Topic: How to use containers without losing speed

Dale Black (Feb 09 2021 at 01:11):

What is the best way to loop through data and store it in an array without sacrificing performance? The code below is what I am working on but I am guessing it is not very efficient

start_bounds = []
end_bounds = []
for i in 1:N
    start_bounds = map(d -> minimum(getindex.(idx, d)), 1:N)
    end_bounds = map(d -> maximum(getindex.(idx, d)), 1:N)
end

If it helps, here is the full function

function generate_spatial_bounding_box(img, select_function)
    idx = findall(select_function, img)
    N = _ndims(eltype(idx))
    start_bounds = []
    end_bounds = []
    for i in 1:N
        start_bounds = map(d -> minimum(getindex.(idx, d)), 1:N)
        end_bounds = map(d -> maximum(getindex.(idx, d)), 1:N)
    end
    return start_bounds, end_bounds
end

Dale Black (Feb 09 2021 at 06:05):

Oh wait, I think I can just avoid the empty array container altogether. This seems to work fine.

function generate_spatial_bounding_box(img, select_function)
    idx = findall(select_function, img)
    N = _ndims(eltype(idx))
    start_bounds = map(d -> minimum(getindex.(idx, d)), 1:N)
    end_bounds = map(d -> maximum(getindex.(idx, d)), 1:N)
    return start_bounds, end_bounds
end

Júlio Hoffimann (Feb 09 2021 at 06:49):

BTW you may take a look at GeoStats.jl if you're interested in bounding boxes of spatial data. Part of this functionality is being moved to Meshes.jl in the next release.

Dale Black (Feb 09 2021 at 17:02):

Meshes.jl is extremely interesting but I work with medical data, so 3D voxels. I’m not sure if meshes will apply to my use case but you might be able to convince me otherwise if there is a benefit that I’m not seeing?

Júlio Hoffimann (Feb 09 2021 at 17:19):

Meshes.jl should be N-dimensional, we support 3D models and that is definitely part of the goal.

Júlio Hoffimann (Feb 09 2021 at 17:20):

You can follow the #meshes.jl stream in case the works overlaps.

Dale Black (Feb 09 2021 at 23:11):

Will do, thanks!

Last updated: Oct 02 2023 at 04:34 UTC