Fold up a matrix!

Question

Given a matrix, sum its values up/down or left/right to form an X, fold it up, and return the list. I describe the algorithm here:

Algorithm

Your input will be a odd-sized square matrix of integers within your language's reasonable numerical capacity.

Let's take the following matrix as an example:

1 2 3 2 1
0 3 2 3 0
4 2 5 6 3
7 4 7 9 4
0 6 7 2 5

First, add every number to the closest number that is on the main diagonal or antidiagonal. That is, divide the matrix into four sections along the main diagonal and antidiagonal, and then sum all of the numbers in each section towards the center, like so:

1   2   3   2   1
    ↓   ↓   ↓    
0 → 3   2   3 ← 0
        ↓        
4 → 2 → 5 ← 6 ← 3
        ↑        
7 → 4   7   9 ← 4
    ↑   ↑   ↑    
0   6   7   2   5

This step gives the following result:

1        1
  5    5
    39
  17  15
0        5

Then, we fold it by flattening the X and interweaving the elements with the top-left first and the bottom left last. This gives the following result:

1, 0, 5, 17, 39, 5, 15, 1, 5

You can imagine this as stretching the main diagonal and rotating it counterclockwise.

This is the final result.

Challenge

Implement this algorithm. Standard loopholes apply. All reasonable I/O formats acceptable.

Test Cases

Input
Output

1 2 3 2 1
0 3 2 3 0
4 2 5 6 3
7 4 7 9 4
0 6 7 2 5

1, 0, 5, 17, 39, 5, 15, 1, 5

1 2 3 4 5
5 4 3 2 1
1 3 5 7 9
0 9 8 7 6
6 7 8 9 0

1, 6, 11, 16, 47, 7, 22, 5, 0

1 3 7 4 8 5 3
8 4 7 5 3 8 0
0 6 3 6 9 8 4
2 6 5 8 7 4 2
0 6 4 3 2 7 5
0 6 7 8 5 7 4
8 5 3 2 6 7 9

1, 8, 15, 11, 23, 20, 62, 32, 25, 13, 18, 3, 9

Answer 1

JavaScript, 113 bytes

s=>(l=s.length-1,a=[],s.map((v,y)=>v.map((n,x)=>a[q=2*[x,y,l-y].sort((u,v)=>u-v)[1]+(y>l/2),q-=q>l]=~~a[q]+n)),a)

f=

s=>(l=s.length-1,a=[],s.map((v,y)=>v.map((n,x)=>a[q=2*[x,y,l-y].sort((u,v)=>u-v)[1]+(y>l/2),q-=q>l]=~~a[q]+n)),a)

<p><textarea id="i" style="width: 400px; height: 120px">1 2 3 2 1
0 3 2 3 0
4 2 5 6 3
7 4 7 9 4
0 6 7 2 5</textarea></p>
<p><button onclick="o.value=f(i.value.trim().split`\n`.map(l=>l.trim().split(/\s+/).map(v => parseInt(v,10)))).join` `">Run</button></p>
<p><output id="o"></output></p>

Answer 2

Python, 159 158 bytes

def f(m):l=len(m)-1;r=range(1,l);return m[0][::l]+f([[sum(m[x][1%y*y:(y>l-2)-~y])+m[0][y]*(x<2)+m[l][y]*(x>l-2)for y in r]for x in r])+m[l][::l]if l else m[0]

Try it online!

Answer 3

Jelly, 25 23 21 bytes

AṂ×ṠṚ
LHŒRṗ2Ç€ḅLĠịFS€

Try it online!

Alternate version, 19 bytes

AṂ×ṠṚ
LHŒRṗ2Ç€ĠịFS€

This didn't use to work because Ġ behaved improperly for nested arrays. The only difference is that the pairs [q, p] mentioned in How it works are sorted lexicographically instead of mapping them to p + nq before sorting.

Try it online!

Background

We start by replacing its elements with coordinates, increasing leftwards and downwards and placing (0, 0) in the center of the matrix.

For a 7x7 matrix M, we get the following coordinates.

(-3,-3) (-3,-2) (-3,-1) (-3, 0) (-3, 1) (-3, 2) (-3, 3)
(-2,-3) (-2,-2) (-2,-1) (-2, 0) (-2, 1) (-2, 2) (-2, 3)
(-1,-3) (-1,-2) (-1,-1) (-1, 0) (-1, 1) (-1, 2) (-1, 3)
( 0,-3) ( 0,-2) ( 0,-1) ( 0, 0) ( 0, 1) ( 0, 2) ( 0, 3)
( 1,-3) ( 1,-2) ( 1,-1) ( 1, 0) ( 1, 1) ( 1, 2) ( 1, 3)
( 2,-3) ( 2,-2) ( 2,-1) ( 2, 0) ( 2, 1) ( 2, 2) ( 2, 3)
( 3,-3) ( 3,-2) ( 3,-1) ( 3, 0) ( 3, 1) ( 3, 2) ( 3, 3)

We now compute the minimum absolute value of each coordinate pair and multiply the signs of both coordinate by it, mapping (i, j) to (sign(i)m, sign(j)m), where m = min(|i|, |j|).

(-3,-3) (-2,-2) (-1,-1) ( 0, 0) (-1, 1) (-2, 2) (-3, 3)
(-2,-2) (-2,-2) (-1,-1) ( 0, 0) (-1, 1) (-2, 2) (-2, 2)
(-1,-1) (-1,-1) (-1,-1) ( 0, 0) (-1, 1) (-1, 1) (-1, 1)
( 0, 0) ( 0, 0) ( 0, 0) ( 0, 0) ( 0, 0) ( 0, 0) ( 0, 0)
( 1,-1) ( 1,-1) ( 1,-1) ( 0, 0) ( 1, 1) ( 1, 1) ( 1, 1)
( 2,-2) ( 2,-2) ( 1,-1) ( 0, 0) ( 1, 1) ( 2, 2) ( 2, 2)
( 3,-3) ( 2,-2) ( 1,-1) ( 0, 0) ( 1, 1) ( 2, 2) ( 3, 3)

Matrix elements that correspond to the same pair have to be summed. To determine the order of the sums, we map each pair (p, q) to p + nq, where n is the number of rows/columns of M.

-24 -16  -8   0   6  12  18
-16 -16  -8   0   6  12  12
 -8  -8  -8   0   6   6   6
  0   0   0   0   0   0   0
 -6  -6  -6   0   8   8   8
-12 -12  -6   0   8  16  16
-18 -12  -6   0   8  16  24

The order of sums corresponds to the order of integers the correspond to its summands.

How it works

LHŒRṗ2Ç€ḅLĠịFS€  Main link. Argument: M (matrix)

L                Compute n, the length (number of rows) of M.
 H               Halve it.
  ŒR             Symmetric range; map t to [-int(t), ..., 0, int(t)].
    ṗ2           Compute the second Cartesian power, yielding all pairs [i, j]
                 with -t ≤ i, j ≤ t.
      ǀ         Map the helper link over the resulting array of pairs.
         L       Yield n.
        ḅ        Unbase; map each pair [q, p] to (p + nq).
          Ġ      Group the indices of the resulting array of n² integers by their
                 corresponding values, ordering the groups by the values.
            F    Flatten M.
           ị     Index into the serialized matrix.
             S€  Compute the sum of each group.


AṂ×ṠṚ            Helper link. Argument: [i, j] (index pair)

A                Absolute value; yield [|i|, |j|].
 Ṃ               Minimum; yield m := min(|i|, |j|).
   Ṡ             Sign; yield [sign(i), sign(j)].
  ×              Multiply; yield [p, q] := [sign(i)m, sign(j)m].
    Ṛ            Reverse; yield [q, p].

Answer 4

Dyalog APL, 101 99 64 62 59 bytes

_{3 bytes saved by @Adám}

{+/∘,¨⍵∘רd∘=¨k[⍋k←↑∪/,d←∘.((+/1x×××⌊/∘|),)⍨(⍳x)-⌈.5×x←≢⍵]}

Try it online!

Using Dennis' awesome algorithm.

Answer 5

APL (Dyalog), 60 bytes*

In collaboration with my colleague Marshall.

Anonymous prefix lambda. Takes matrix as argument and returns vector. Assumes ⎕IO (Index Origin) to be zero, which is default on many systems.

{(,⍉{+/,(s×-×⍺)↓⍵×i∊¨⍨s←⌊⊃r÷2}⌺r⊢⍵)/⍨,(⊢∨⌽)=/¨i←⍳r←⍴⍵}

Try it online!

{…} anonymous lambda; ⍵ is right argument (as rightmost letter of the Greek alphabet):

 ⍴⍵ shape of the argument (list of two identical elements)

 r← store as r (as in rho)

 ⍳ all ɩndices of an array of that size, i.e. (0 0),(0 1)…

 i← store in i (as in iota)

 =/¨ Boolean where the coordinates are equal (i.e. the diagonal)

 (…) apply this anonymous tacit prefix function:

  ⌽ reverse the argument

  ⊢∨ OR that with the unmodified argument

 , ravel (straighten into simple list)

 We now have a Boolean mask for the diagonals.

 (…)/⍨ use that to filter the following:

  ⊢⍵ yield (to separate from r) the argument

  {…}⌺r call the following anonymous infix lambda on each element, with the r-neighbourhood (padded with zeros as needed) as right argument (⍵), and a two element list of number padded rows,columns (negative for bottom/right, zero for none) as left argument (⍺):

   r÷2 divide r with two

   ⊃ pick the first element (they are identical)

   ⌊ floor it

   s← store as s (for shape)

   i∊⍨¨ for each element of i, Boolean if s is a member thereof

   ⍵× multiply the neighbourhood therewith

   (…)↓ drop the following number of rows and columns (negative for bottom/right):

    ×⍺ signum of the left argument (i.e. the direction of paddings)

    - negate

    s× multiply s therewith

   , ravel (straighten into list)

   +/ sum (plus reduction)

Now we have full matrix of sums, but we need to filter all the values read columnwise.

  ⍉ transpose

  , ravel (straighten into simple list)

---

* By counting ⌺ as ⎕U233A . Try it online!