Vectorized index generation based on multiple conditions

Hi all,

I have a huge column vector from which I need to extract specific values using vectorization instead of loops. The algorithm is like this:

Take first 5 values, check whether first or second value is greater than 48. If it is, remove those five values from the vector. Else remove first two values from the vector. Then take next five values and repeat.

This is the code based on loop:

data = randi([1 100],10000,1);
ind = ones(length(data),1);
for i = 1:5:length(data)
    if (data(i) > 48 || data(i+1) > 48)
        ind(i:i+4) = 0;
    else
        ind(i:i+1) = 0;
    end
end
data_trunc = data(logical(ind));

What would the most effcient vectorized method to parse my vector?

Thanks

[Turlough Hughes]

I'm sure one could do better but the following is defintely faster:

data = randi([1 100],10000,1);

data_trunc=reshape(data,[5,length(data)/5]); % convert to 5 by 2000 array

index_a=data_trunc(1,:)>48;
index_b=data_trunc(2,:)>48;
data_trunc(1:2,:)=[];           % delete rows 1 and 2 as they are no longer needed    

c=or(index_a,index_b);    % logical row index 
c=[c;c;c];                % logic array to delete values as required
data_trunc(c)=[];

Running the following code shows the difference in speed:

data = randi([1 100],10000,1);
tic
for ii=1:10000
    data_trunc=reshape(data,[5,length(data)/5]); % convert to 5 by 2000 array

    index_a=data_trunc(1,:)>48;
    index_b=data_trunc(2,:)>48;
    data_trunc(1:2,:)=[];
    
    c=or(index_a,index_b);
    c=[c;c;c];  
    data_trunc(c)=[];
end
toc

On my computer it took 0.778 seconds to run it 10000 times.

tic
for ii=1:10000
    ind = ones(length(data),1);
    for i = 1:5:length(data)
        if (data(i) > 48 || data(i+1) > 48)
            ind(i:i+4) = 0;
        else
            ind(i:i+1) = 0;
        end
    end
    data_trunc = data(logical(ind));
end
toc

Whereas implementing same using your looped example takes 16.7 seconds.