Looping in a spiral

Question

A friend was in need of an algorithm that would let him loop through the elements of an NxM matrix (N and M are odd). I came up with a solution, but I wanted to see if my fe

Answer 1

Here's an answer in Julia: my approach is to assign the points in concentric squares ('spirals') around the origin (0,0), where each square has side length m = 2n + 1, to produce an ordered dictionary with location numbers (starting from 1 for the origin) as keys and the corresponding coordinate as value.

Since the maximum location per spiral is at (n,-n), the rest of the points can be found by simply working backward from this point, i.e. from the bottom right corner by m-1 units, then repeating for the perpendicular 3 segments of m-1 units.

This process is written in reverse order below, corresponding to how the spiral proceeds rather than this reverse counting process, i.e. the ra [right ascending] segment is decremented by 3(m+1), then la [left ascending] by 2(m+1), and so on - hopefully this is self-explanatory.

import DataStructures: OrderedDict, merge

function spiral(loc::Int)
    s = sqrt(loc-1) |> floor |> Int
    if s % 2 == 0
        s -= 1
    end
    s = (s+1)/2 |> Int
    return s
end

function perimeter(n::Int)
    n > 0 || return OrderedDict([1,[0,0]])
    m = 2n + 1 # width/height of the spiral [square] indexed by n
    # loc_max = m^2
    # loc_min = (2n-1)^2 + 1
    ra = [[m^2-(y+3m-3), [n,n-y]] for y in (m-2):-1:0]
    la = [[m^2-(y+2m-2), [y-n,n]] for y in (m-2):-1:0]
    ld = [[m^2-(y+m-1), [-n,y-n]] for y in (m-2):-1:0]
    rd = [[m^2-y, [n-y,-n]] for y in (m-2):-1:0]
    return OrderedDict(vcat(ra,la,ld,rd))
end

function walk(n)
    cds = OrderedDict(1 => [0,0])
    n > 0 || return cds
    for i in 1:n
        cds = merge(cds, perimeter(i))
    end
    return cds
end

So for your first example, plugging m = 3 into the equation to find n gives n = (5-1)/2 = 2, and walk(2) gives an ordered dictionary of locations to coordinates, which you can turn into just an array of coordinates by accessing the dictionary's vals field:

walk(2)
DataStructures.OrderedDict{Any,Any} with 25 entries:
  1  => [0,0]
  2  => [1,0]
  3  => [1,1]
  4  => [0,1]
  ⋮  => ⋮

[(co[1],co[2]) for co in walk(2).vals]
25-element Array{Tuple{Int64,Int64},1}:
 (0,0)  
 (1,0)  
 ⋮       
 (1,-2) 
 (2,-2)

Note that for some functions [e.g. norm] it can be preferable to leave the coordinates in arrays rather than Tuple{Int,Int}, but here I change them into tuples—(x,y)—as requested, using list comprehension.

The context for "supporting" a non-square matrix isn't specified (note that this solution still calculates the off-grid values), but if you want to filter to only the range x by y (here for x=5,y=3) after calculating the full spiral then intersect this matrix against the values from walk.

grid = [[x,y] for x in -2:2, y in -1:1]
5×3 Array{Array{Int64,1},2}:
 [-2,-1]  [-2,0]  [-2,1]
   ⋮       ⋮       ⋮ 
 [2,-1]   [2,0]   [2,1]

[(co[1],co[2]) for co in intersect(walk(2).vals, grid)]
15-element Array{Tuple{Int64,Int64},1}:
 (0,0)  
 (1,0)  
 ⋮ 
 (-2,0) 
 (-2,-1)