Array order for maximum pair matches

Question

I have an array:

array([[ 4, 10],
       [ 4,  2],
       [ 0,  7],
       [ 5, 11],
       [ 6,  8],
       [ 3,  6],
       [ 9,  7],
       [ 2, 11],
       [ 9,  5],
       [ 8,  1]])

I need a method with which I can arrange pairs of values ​​so that as many paired two-element sets as possible have a common value. This is an example of a desired ordered array:

array([[ 4, 10],
       [ 4,  2],
       [ 2, 11],
       [ 5, 11],
       [ 9,  5],
       [ 9,  7],
       [ 0,  7],  #note the gap here:
       [ 8,  1],
       [ 6,  8],
       [ 3,  6]])

There are several conditions in these arrays. No duplicate pairs (ie, [1,0] and [0,1] will appear elsewhere in the array if [0,1] already exists). No pair has the same value (ie: [1,1] will not exist). No pair will have more than two matches (iow: no value more than two times in the whole array), but a pair can only have zero matches (note that in the above array there is a space for which there is no match).

, , . , - , . , : 1) , numpy collections . 2) , numpy .sort() ( ), . 3) , , , . ( ", -!" )

, :

def shuffled(N=12, ans=1):
    '''returns is now the unsorted test array'''
    r = range(N)
    random.shuffle(r)
    l = []
    for i in range(N):
        l.append((r[i-1], r[i]))
    random.shuffle(l)
    return np.array(l)[ans+1:]

# step 2: ???

def test_ordered(a):
    '''checks if generated array has been sorted'''
    c0 = a[1:,0]==a[:-1,0]
    c1 = a[1:,0]==a[:-1,1]
    c2 = a[1:,1]==a[:-1,0]
    c3 = a[1:,1]==a[:-1,1]
    cond = c0+c1+c2+c3
    ans = sum(numpy.logical_not(cond))
    # when sorted this should return the same number input into
    # shuffled() as 'ans':
    return ans

( ? , .)

:

. 20% , , , , , . , "" "". . . 10 000 N. 0,5 . z - .

!

Answer 1

, - , - .

, . , ( ). :

• [ ] , 1 ( ). , , .
• [ ] 1 , , . ( 2). -, , , .
• , .

: 1 2. ( ), : 1 2, 1. , .

: dict .

, (, O (1) dict).

.

import collections

def handle_vertex(v, vs):
  if v in vs[0]:
    vs[0].remove(v)
  elif v in vs[1]:
    vs[0].add(v)
    vs[1].remove(v)

def follow(edgemap, start, vs):
  """Follow a path in the graph, yielding the edges."""
  v0 = start
  last = None
  while True:
    # All vertices that we can go to next.
    next_v = [v for v in edgemap[v0] if v != last]
    if not next_v:
      # We've reached the end (we must have been a line).
      return
    assert len(next_v) == 1 or (v0 == start and len(next_v) == 2)
    next_v = next_v[0]
    # Remove the endpoints from the vertex-degree sets.
    handle_vertex(v0, vs)
    handle_vertex(next_v, vs)
    yield v0, next_v
    if next_v == start:
      # We've got back to the start (we must have been a cycle).
      return
    v0, last = next_v, v0

def pairsort(edges):
  edgemap = collections.defaultdict(list)
  original_edges = {}
  for a, b in edges:
    # Build the adjacency table.
    edgemap[a].append(b)
    edgemap[b].append(a)
    # Keep a map to remember the original order pairs appeared in
    # so we can output edges correctly no matter which way round
    # we store them.
    original_edges[a, b] = [a, b]
    original_edges[b, a] = [a, b]
  # Build sets of degree 1 and degree 2 vertices.
  vs = [set(), set()]
  for k, v in edgemap.iteritems():
    vs[len(v) - 1].add(k)
  # Find all components that are lines.
  while vs[0]:
    v0 = vs[0].pop()
    for e in follow(edgemap, v0, vs):
      yield original_edges[e]
  # Find all components that are cycles.
  while vs[1]:
    v0 = vs[1].pop()
    for e in follow(edgemap, v0, vs):
      yield original_edges[e]

input = [
    [ 4, 10],
    [ 4,  2],
    [ 0,  7],
    [ 5, 11],
    [ 6,  8],
    [ 3,  6],
    [ 9,  7],
    [ 2, 11],
    [ 9,  5],
    [ 8,  1]]

print list(pairsort(input))

Answer 2

, ; , , .

: 1D- , , . NumPy roll, (-), , , .

0s. (tx, ), , , NP.argmax(tx).

import numpy as NP

# create some data:
c1 = NP.random.randint(0, 10, 15)
c2 = NP.random.randint(0, 10, 15)
c12 = NP.concatenate((c1, c2)).reshape(15, 2)

tx = []    # to hold the indices of the various orderings

for i in range(15) :
    v = NP.diff(c12, axis=0)
    num_equal_neighbors = NP.sum( NP.sum(v==0, axis=0) )
    tx.append(num_equal_neighbors)
    c2 = NP.roll(c2, 1)
    c12[:,1] = c2

, "" :

best_order = NP.argmax(tx)

, 1D * best_order * ( )

Answer 3

Here is a simpler implementation of basically the same algorithm described in Paul Hankin's answer using different data structures. It also works in linear time.

edges = [[4, 10], [4,  2], [0,  7], [5, 11], [6,  8],
         [3,  6], [9,  7], [2, 11], [9,  5], [8,  1]]

def add_edge(vertex, adj):
    edge = edges[adj[0]][:]
    ordered_edges.append(edge[:])
    edge.remove(vertex)
    new_vertex = edge[0]
    new_adj = adj_edges[new_vertex]
    new_adj.remove(adj[0])
    del adj[0]
    return new_vertex, new_adj

adj_edges = {}
for i, edge in enumerate(edges):
    for vertex in edge:
        adj_edges.setdefault(vertex, []).append(i)
ordered_edges = []
for vertex, adj in adj_edges.iteritems():
    while len(adj) == 1:
        vertex, adj = add_edge(vertex, adj)
for vertex, adj in adj_edges.iteritems():
    while adj:
        vertex, adj = add_edge(vertex, adj)

Answer 4

See also The longest path problem : NP complete for general graphs or shortest path with weights negative for acyclic.
Also try the graph shown in Cladding Tree : the longest is harder than just the longest.