How to get the sum of numbers below a desired number in sets?

How to get the sum of the numbers in the array which are below -1 in sets.

For example

a=[1 -2 0 .5 -3 -4 5 7 -2 -3 -4];

so here the numbers which are consecutively less than -1 should be added only.

i.e. The answer should be

b=[-2 -7 -10]

PS: Actual data is very big.

Answer 1

[Image Analyst]

These two lines of code will do it.

a = [1 -2 0 .5 -3 -4 5 7 -2 -3 -4];
props = regionprops(a < -1, a, 'Area', 'MeanIntensity');
output = [props.Area] .* [props.MeanIntensity]

How big are your arrays? This should be pretty speedy for up to tens or hundreds of millions of elements. If you have gigabyte sized arrays, it could take a few minutes. The code requires the Image Processing Toolbox.

Answer 2

[per isakson]

"data is very big" How big is that? Does it fit in the memory?

Test this function

function  b = cssm( a )
    %{
    b = cssm([-2,0,.5,-3,-4,5,7,-2,-3,-4]);
    b = cssm([1,-2,0,.5,-3,-4,5,7,-2,-3,-4]);
    b = cssm([1,-2,0,.5,-3,-4,5,7,-2,-3,-4,3]);
    a = randn(1,1e6);
    tic, b = cssm( a ); toc
    %}
    %%
    b = zeros(size(a));
    %%
    kk = 1;
    for jj = find(a<-1,1,'first') : length(a)
        if a(jj) < -1
            b(kk) = b(kk) + a(jj);
        else
            if b(kk) ~= 0
                kk = kk+1;
            end
        end
    end
    if a(end) < -1
        b( kk+1 : end ) = [];
    else
        b( kk : end ) = [];
    end
end

Performance comparison with the solution of Image Analyst

The test below indicates that the solution using a for-loop is significantly faster - in this case.

>> a = randn(1,1e6);
>> [ isEQ, et ] = cssm( a )
isEQ =
  logical
   1
et =
    0.0069    1.0598
    
>> a = randi([-12,12],1,1e6,'int8');
>> [ isEQ, et ] = cssm( a )
isEQ =
  logical
   1
et =
    0.0138    3.5893    

where

function  [ isEQ, et ] = cssm( a )
    tic
    b1 = loop( a );
    et(1) = toc;
    tic
    b2 = ia( a );
    et(2) = toc;
    isEQ = all(abs(b1-b2)<1e-6);
end
function    b = loop( a )
    %%
    b = zeros(size(a));
    %%
    kk = 1;
    for jj = find(a<-1,1,'first') : length(a)
        if a(jj) < -1
            b(kk) = b(kk) + a(jj);
        else
            if b(kk) ~= 0
                kk = kk+1;
            end
        end
    end
    if a(end) < -1
        b( kk+1 : end ) = [];
    else
        b( kk : end ) = [];
    end
end
function    output = ia( a )
    props = regionprops(a < -1, a, 'Area', 'MeanIntensity');
    output = [props.Area] .* [props.MeanIntensity];    
end

Caveat: It happens that I maketakes.