A very fast way to find elements and their indices? (Is ismember fast?)

A very fast way to find elements and their indices? (Is ismember fast?)

This would be my example:

% input: create arrays "a" and "b"
a = {'A23'};
l = 'A' : 'Z';
rl = reshape(l,size(l,2),1);
for i = 1 : size(rl,1)
    for j = 1 : 10000
           f{i,j} = sprintf('%s%d',rl(i), j);
    end
end
b = reshape(f,[],1);

% still part of the input: randomly replace 50 values of "b" with {'A23'}
r = round((size(b,1)-1) .* rand(50,1) + 1);
b(r) = a;


% How can I make this part way much faster ?
tic
[~,idx2] = ismember(string(b),string(a));
b_filtered = b(find(idx2),:);
toc

Elapsed time is 0.057298 seconds.

[Stephen23]

https://undocumentedmatlab.com/articles/ismembc-undocumented-helper-function

[John D'Errico]

Is ismember fast? Well, yes. Pretty fast. Is it fast enough? Of course not. There will always be somebody who expects everything in MATLAB to operate within a femtosecond on an arbitrarily large array.

I might note, just looking at your code, that this does NOT choose integers uniformly in the interval you think it does:

% still part of the input: randomly replace 50 values of "b" with {'A23'}
r = round((size(b,1)-1) .* rand(50,1) + 1);

The first and last numbers, thus here 1 and size(b,1)-1, will be generated 50% as often as the rest of the population. So those two will be under-represented. Your code though.

[Steven Lord]

If you're trying to generate random integer values in an interval, use randi instead of rand.

[Sim]

Thanks a lot @Stephen23, @John D'Errico, @Steven Lord !!!

@Stephen23, yes, I can try the "ismembc" function :-) .....I hope it is "safe", without "side effects" :-)

@John D'Errico, yes, thank you, I have messed up the random part... :-)

@Steven Lord, yes, I will go for "randi", thanks! :-)

Answer 1

[Bruno Luong]

You overly complicate your code for nothing, and yes ismember if fast.

Not sure if your a is always single element or just in this example.

% input: create arrays "a" and "b"
a = {'A23'};
l = 'A' : 'Z';
rl = reshape(l,size(l,2),1);
for i = 1 : size(rl,1)
    for j = 1 : 10000
        f{i,j} = sprintf('%s%d',rl(i), j);
    end
end
b = reshape(f,[],1);

% still part of the input: randomly replace 50 values of "b" with {'A23'}
r = round((size(b,1)-1) .* rand(50,1) + 1);
b(r) = a;


% How can I make this part way much faster ?
tic
[~,idx2] = ismember(string(b),string(a));
b_filtered = b(find(idx2),:);
toc

Elapsed time is 0.062962 seconds.

tic
tf = ismember(b,a);
b_filtered = b(tf,:);
toc

Elapsed time is 0.009279 seconds.

% only when a is scalar
tic
b_filtered = b(strcmp(b, a{1}));
toc

Elapsed time is 0.003851 seconds.

[Sim]

Thanks a lot @Bruno Luong, it is around 7 times faster :-)

Question... do you think I can use ismembc as well ?

tf = ismembc(b,a);

Could it be even faster ?

If Yes, would that function be recommended ? :-)

[Bruno Luong]

AFAIK ismembc work on numbers, not char array or string, and the second argumentr must be sorted. So it is NOT applicable in your case.

Personally I know this function but I never use it since it is undocumented and I never need to draw the last ounce of speed for ismember.

[Sim]

Thanks a lot @Bruno Luong!! :-)

About: "Not sure if your a is always single element or just in this example.", Yes, it is always a single element (that change in a loop, but always single).