途径
使用arrayfun和ismember –
Result = A(arrayfun(@(n) any(ismember(B,A{n})),1:numel(A)))
或者使用arrayfun和bsxfun –
Result = A(arrayfun(@(n) any(any(bsxfun(@eq,B(:),A{n}),2)),1:numel(A)))
或者使用arrayfun和setdiff –
Result = A(arrayfun(@(n) numel(setdiff(B,A{n})) < numel(B),1:numel(A)))
或者使用arrayfun和intersect –
Result = A(arrayfun(@(n) ~isempty(intersect(B,A{n})),1:numel(A)))
人们也可以在这里使用cellfun,这样四个基于cellfun的对应解决方案最终会像这样 –
Result = A(cellfun(@(x) any(ismember(B,x)), A))
Result = A(cellfun(@(x) any(any(bsxfun(@eq,B(:),x),2)),A))
Result = A(cellfun(@(x) numel(setdiff(B,x)) < numel(B),A))
Result = A(cellfun(@(x) ~isempty(intersect(B,x)),A))
采取不同的路线[使用bsxfun的掩蔽能力]
不是进入那些基本上是循环方法的基于arrayfun或cellfun的方法,而是可以通过将A转换为2D数值阵列来使解决方案很多.因此,这里的想法是有一个2D数组,其中行数是A中的最大元素数,列数是A中的单元数.该数组的每一列都包含来自A和NaN的每个单元的元素会填满空地.
使用这种方法的解决方案代码看起来像这样 –
lens = cellfun('length',A); %// number of elements in each cell of A
mask = bsxfun(@ge,lens,(1:max(lens))'); %//'# mask of valid places in the 2D array
A_arr = NaN(size(mask)); %//initialize 2D array in which A elements are to be put
A_arr(mask) = [A{:}]; %// put the elements from A
%// Find if any element from B is in any element along the row or dim-3
%// locations in A_arr. Then logically index into A with it for the final
%// cell array output
Result = A(any(any(bsxfun(@eq,A_arr,permute(B,[1 3 2])),1),3));
验证
>> celldisp(Result)
Result{1} =
2 5
Result{2} =
33 77 4
Result{3} =
10 66 17 7 8 11
Result{4} =
15 41 88
标杆
对于有兴趣了解运行时性能的人来说,这是一个快速的基准测试,具有足够大的数据量 –
%// Create inputs
N = 10000; %// datasize
max_num_ele = 100; %// max elements in any cell of A
num_ele = randi(max_num_ele,N,1); %// number of elements in each cell of A
A = arrayfun(@(n) randperm(N,num_ele(n)), 1:N, 'uni', 0);
B = randperm(N,num_ele(1));
%// Warm up tic/toc.
for k = 1:100000
tic(); elapsed = toc();
end
%// Start timing all approaches
disp('************************ With arrayfun **************************')
disp('------------------------ With arrayfun + ismember')
tic
Result = A(arrayfun(@(n) any(ismember(B,A{n})),1:numel(A)));
toc, clear Result
disp('------------------------ With arrayfun + bsxfun')
tic
Result = A(arrayfun(@(n) any(any(bsxfun(@eq,B(:),A{n}),2)),1:numel(A)));
toc, clear Result
disp('------------------------ With arrayfun + setdiff')
tic
Result = A(arrayfun(@(n) numel(setdiff(B,A{n})) < numel(B),1:numel(A)));
toc, clear Result
disp('------------------------ With arrayfun + intersect')
tic
Result = A(arrayfun(@(n) ~isempty(intersect(B,A{n})),1:numel(A)));
toc, clear Result
disp('************************ With cellfun **************************')
disp('------------------------ With cellfun + ismember')
tic
Result = A(cellfun(@(x)any(ismember(B,x)), A));
toc, clear Result
disp('------------------------ With cellfun + bsxfun')
tic
Result = A(cellfun(@(x) any(any(bsxfun(@eq,B(:),x),2)),A));
toc, clear Result
disp('------------------------ With cellfun + setdiff')
tic
Result = A(cellfun(@(x) numel(setdiff(B,x)) < numel(B),A));
toc, clear Result
disp('------------------------ With cellfun + setdiff')
tic
Result = A(cellfun(@(x) ~isempty(intersect(B,x)),A));
disp('************************ With masking bsxfun **************************')
tic
lens = cellfun('length',A); %// number of elements in each cell of A
mask = bsxfun(@ge,lens,(1:max(lens))'); %//'
A_numarr = NaN(size(mask));
A_numarr(mask) = [A{:}];
Result = A(any(any(bsxfun(@eq,A_numarr,permute(B,[1 3 2])),1),3));
toc
在我的系统上获得的结果是 –
************************ With arrayfun **************************
------------------------ With arrayfun + ismember
Elapsed time is 0.409810 seconds.
------------------------ With arrayfun + bsxfun
Elapsed time is 0.157327 seconds.
------------------------ With arrayfun + setdiff
Elapsed time is 1.154602 seconds.
------------------------ With arrayfun + intersect
Elapsed time is 1.081729 seconds.
************************ With cellfun **************************
------------------------ With cellfun + ismember
Elapsed time is 0.392375 seconds.
------------------------ With cellfun + bsxfun
Elapsed time is 0.143341 seconds.
------------------------ With cellfun + setdiff
Elapsed time is 1.101331 seconds.
------------------------ With cellfun + setdiff
************************ With masking bsxfun ********************
Elapsed time is 0.067224 seconds.
可以看出,基于cellfun的解决方案比基于arrayfun的解决方案快一点!此外,基于掩码的bsxfun方法看起来很有趣,但请记住它的内存饥饿性.