How to average only positive values in an array?

My problem: I have a matrix M with 20 columns and thousands of rows. Each column in the matrix has positive and negative values (and each column has a different ratio of positive to negative numbers). I want to find the average value of only the positive numbers in each column in the matrix, and put those averages into an array (so a 1 x 20 array). How can I code this?

Answer 1

[Jan]

x = rand(1000, 20) - 0.5;
Positive = x > 0;
x(~Positive) = 0;
MeanPositive = sum(x, 2) ./ sum(Positive, 2);

[Ute123]

This almost did it. I wanted the average of the columns, not the rows. I had to look up the sum function in Matlab help, the 2nd argument needs to be 1, not 2. So I just changed your last line to: MeanPositive = sum(x, 1) ./ sum(Positive, 1);, and that worked.

Thanks!

Answer 2

[Walter Roberson]

arrayfun(@(K) mean(M(:,M(:,K)>0)), 1:size(M,2))

[Ute123]

It didn't work for me, I must be missing something. I get an error that says, "??? Undefined function or method 'arrayfun' for input arguments of type 'function_handle'." Do you know what I need to fix? Thanks.

[Jan]

@Ute: Which Matlab version are you using?

[Ute123]

I'm using version 7.9 (R2009b)

[Walter Roberson]

I did have an error: it should have been

arrayfun(@(K) mean(M(M(:,K)>0),K), 1:size(M,2))

However, if your R2009B does not have arrayfun over function handles then something is wrong with it. The documentation for that version is

http://www.mathworks.com/help/releases/R2009b/techdoc/ref/arrayfun.html

Notice that according to the documentation the "fun" parameter is a function handle.

There may have been versions of arrayfun too old to support function handles, but anything closer to the 2009 releases support them for arrayfun()

Answer 3

[Derek O'Connor]

At the risk of offending the "vectorizers", the function below is 2 to 3 times faster than Jan's answer:

function MeanP = MeanPos(A)
% Calculate the mean of the positive values in each column of A.
% Fast if m (rows) >> n (cols). Slow if m << n.
% Derek O'Connor 19 July 2011. [e-mail]
%
[m,n] = size(A);
MeanP(1:n) = 0;
for j = 1:n
    npos = 0;
    for i = 1:m
        if A(i,j) > 0
            npos = npos+1;
            MeanP(j) = MeanP(j) + A(i,j);
        end
    end
    if npos > 0
        MeanP(j) = MeanP(j)/npos;
    end
end

Here are some run times (secs), with n = 20 cols., m = 10^4, ..., 10^7 rows

         m        Tj        Td      Tj/Td
       -----------------------------------
       10^4     0.0091    0.0044    2.0776

       10^5     0.1356    0.0437    3.1001

       10^6     0.9935    0.4350    2.2839

       10^7     9.9366    4.3983    2.2592
       -----------------------------------

Dell Precision 690, Intel 2xQuad-Core E5345 @ 2.33GHz, 16GB RAM

Windows7 64-bit Prof., MATLAB R2008a, 7.6.0.324

Derek O'Connor