我正在尝试标记一些 numpy 向量(矩阵 M),以便标签显示帕累托边界索引。
例如,非支配向量集 ( v0
) 将用帕累托边界索引 0
标记,非支配向量集 v1 (v1 = M - v0) 将用索引 1
标记,下一个集合/边界非支配向量 v2 (v2 = M - v0 - v1) 2
依此类推,直到矩阵 M 的所有向量都被标记。
我已经将一些测试用例放在一起,但无论我想出什么,要么效率极低(目前不要太在意),要么就是不起作用。
mat1 = np.asarray([
[1, 2, 3, 4, 5, 7],
[1, 2, 3, 4, 5, 6],
[1, 2, 3, 4, 5, 6],
])
calc_fronts(mat1) == [0, 1, 1]
mat2 = np.asarray([
[1, 2, 3, 4, 5, 7],
[1, 2, 3, 4, 5, 6],
[1, 22, 3, 4, 5, 6],
])
calc_fronts(mat2) == [0, 1, 0]
mat3 = np.asarray([
[1, 2, 3, 4, 5, 7],
[1, 2, 3, 4, 5, 5],
[1, 2, 2, 4, 5, 4],
])
calc_fronts(mat3) == [0, 1, 2]
mat4 = np.asarray([
[0, 2, 3, 4, 5, 7],
[1, 2, 3, 4, 5, 6],
[1, 22, 2, 4, 5, 6],
])
calc_fronts(mat4) == [0, 0, 0]
如果 foreach a in x, b in y,向量 x 支配 y:a >= b 并且在 x 中至少存在一个 a > b。
这是我的尝试:
def calc_fronts_2(vecs):
n = vecs.shape[0]
# find the most dominating
m = np.zeros(vecs.shape, dtype=bool)
dominates = []
for i, vec in enumerate(vecs):
for x in xrange(n):
m[x] = i != x
mvecs = np.ma.masked_array(vecs, mask=~m)
# print 'i=', i
# print 'all better \n', np.all(vec >= mvecs, axis=1)
# print 'at least one\n', np.any(np.all(vec > vecs[m], axis=0))
# print 'at least one\n', vec > mvecs
# print 'at least one\n', np.any(vec > mvecs, axis=1)
dom = np.where(np.all(vec >= mvecs, axis=1) & np.any(vec > mvecs, axis=1))
dom = dom[0]
dominates.append(dom.tolist())
# print dominates
dominated_by = [[j for j in xrange(n) if i in dominates[j]] for i in xrange(n)]
print 'domin:\n', dominates
print 'dom by\n', dominated_by
ranks = np.empty(n, dtype=int)
ranks.fill(-1)
for r in xrange(n):
remove = set()
for i in xrange(n):
if ranks[i] == -1 and len(dominated_by[i]) == 0:
ranks[i] = r
remove.add(i)
for ranked in remove:
for domby in dominated_by:
if ranked in domby:
domby.remove(ranked)
if np.all(ranks == -1):
break
return ranks
最佳答案
这是我的大量尝试,希望能提供一些新的想法。但请注意,创建了 (m,m,n) 临时数组(当 M
为 (m,n) 形状时)。
import numpy as np
def calc_fronts(M):
i_dominates_j = np.all(M[:,None] >= M, axis=-1) & np.any(M[:,None] > M, axis=-1)
remaining = np.arange(len(M))
fronts = np.empty(len(M), int)
frontier_index = 0
while remaining.size > 0:
dominated = np.any(i_dominates_j[remaining[:,None], remaining], axis=0)
fronts[remaining[~dominated]] = frontier_index
remaining = remaining[dominated]
frontier_index += 1
return fronts
演示:
In [217]: M = np.array([[ 0, 18, 1],
...: [ 4, 8, 11],
...: [ 3, 19, 3],
...: [18, 1, 19],
...: [16, 7, 13],
...: [ 3, 18, 3],
...: [15, 13, 19],
...: [13, 5, 13],
...: [ 2, 16, 16],
...: [ 9, 14, 17]])
In [218]: calc_fronts(M)
Out[218]: array([2, 1, 0, 0, 0, 1, 0, 1, 0, 0])