我会说这是你迄今为止的一个很好的解决方案。
我看到了一些优化。这是我的变化:
% INITIALIZE
NN = 9;
L = rand(1,NN-1);
while numel(L) ~= NN;
L = unique( randi(100,1,NN) ); end
% Some bogus constraints
constraints = [...
L(1) L(2)
L(3) L(6)
L(3) L(5)
L(8) L(4)];
% METHOD 0 (your original method)
tic
orderings = perms(L);
p = size(orderings,1);
c = size(constraints,1);
toKeep = true(p,1);
for perm = 1:p
for constr = 1:c
idxA = find(orderings(perm,:) == constraints(constr,1));
idxB = find(orderings(perm,:) == constraints(constr,2));
if idxA > idxB
toKeep(perm) = false;
end
end
end
orderings0 = orderings(toKeep,:);
toc
% METHOD 1 (your original, plus a few optimizations)
tic
orderings = perms(L);
p = size(orderings,1);
c = size(constraints,1);
toKeep = true(p,1);
for perm = 1:p
for constr = 1:c
% break on first condition breached
if toKeep(perm)
% find only *first* entry
toKeep(perm) = ...
find(orderings(perm,:) == constraints(constr,1), 1) < ...
find(orderings(perm,:) == constraints(constr,2), 1);
else
break
end
end
end
orderings1 = orderings(toKeep,:);
toc
% METHOD 2
tic
orderings = perms(L);
p = size(orderings,1);
c = size(constraints,1);
toKeep = true(p,1);
for constr = 1:c
% break on first condition breached1
if any(toKeep)
% Vectorized search for constraint values
[i1, j1] = find(orderings == constraints(constr,1));
[i2, j2] = find(orderings == constraints(constr,2));
% sort by rows
[i1, j1i] = sort(i1);
[i2, j2i] = sort(i2);
% Check if columns meet condition
toKeep = toKeep & j1(j1i) < j2(j2i);
else
break
end
end
orderings2 = orderings(toKeep,:);
toc
% Check for equality
all(orderings2(:) == orderings1(:))
结果:
Elapsed time is 17.911469 seconds. % your method
Elapsed time is 10.477549 seconds. % your method + optimizations
Elapsed time is 2.184242 seconds. % vectorized outer loop
ans =
1
ans =
1
然而,整个方法有一个基本缺陷恕我直言;的直接使用perms
。由于内存限制(NN < 10
如 中所述help perms
),这固有地造成了限制。
我强烈怀疑,当您将自定义的perms
. 幸运的是,perms
它不是内置的,因此您可以先将该代码复制粘贴到您的自定义函数中。