how calculate percentage with a vector and a matrix with nan elements?

I have a vector V where each element of V is bigger than zero, and a matrix M where each row represent data, each column the source of the data, and each element the number of errors.

A small example will be:

 V=[10 5 4 8];

 M=[0   2   NaN 4  ;
    0   0   3   2  ;
    0   NaN 0   1  ;
    0   2   NaN 0  ];

I would like to calculate a vector P with the percentage or error without for loops and without consider the part of the matrix with NaN elements such as

 P=[((0+0+0+0)/(10+5+4+8))*100 ((2+0+2)/(10+5+8))*100 ((3+0)/(5+4))*100 ((4+2+1+0)/(10+5+4+8))*100]
 P=[ 0   17.3913   33.3333   25.9259]

Answer 1

[dpb]

>> m=M;m(isnan(m))=0;
>> sum(m)./sum(repmat(V.',1,size(M,2)).*isfinite(M))*100
ans =
       0   17.3913   33.3333   25.9259
>>

If you don't need M any longer, can do the substitution in place, of course. It'd be nice if were syntax with M(isfinite(M)) could return an array with a filler value for the missing locations as an alternate syntax to needing the temporary variable.

ADDENDUM

Oh, forgotted about nansum; it's in Statistics Toolbox in R2012b; not sure if got propagated to base product later or not; seems like maybe??? With it, can get rid of explicit temporary step...

nansum(M)./sum(repmat(V.',1,size(M,2)).*isfinite(M))*100