Error using 'omitnan'

Hi MATLAB community,

I wrote a code for identify 999999 to substitute to NaN, my purpose is replace NaN to mean of column data,

I tried follow code:

[l,c]=size(matrix) % size (324,25)
for w=1:c
   for q = 1:l
   if matrix(q,w)==999999     
      matrix(q,w) =NaN;       
      matrix(isnan(matrix))= mean(matrix(:,w),'omitnan'); % in the case NaN is present in column 21, it will substitute the element  for mean of all element in column 21
   end
   end
end   

Could help me?

Grateful

[Walter Roberson]

Omitnan did not exist in your release. One of the toolboxes had nanmean()

Answer 1

[Adam Danz]

Check out nanmean() in the stats & machine learning toolbox.

An alternative that doesn't require any toolboxes:

nanidx = isnan(matrix(:,w)); 
mean(matrix(~nanidx,w)

Update following comments below

To compute the mean of each column of a matrix while ignoring a key value (999999),

% Create fake data
key = 999999; 
matrix = magic(10); 
matrix(randi(100,1,10)) = key; 

% Identify the columns that contain at least one 999999
isKey = matrix == key; 
colIdx = any(isKey,1); 
% Count the number of rows per column that are not 999999
rowCount = sum(~isKey); 
% Temporarily replace 999999 with 0 and calculate the column means
matrixTemp = matrix .* ~isKey; 
colMean = sum(matrixTemp)./rowCount;

% The 2nd block above can be reduced to
%    isKey = matrix == key; 
%    colMean = sum(matrix .* ~isKey)./ sum(~isKey);

[Guilherme Lopes de Campos]

Thank you very much Adam, for help,

It's correct how I inserted?

for w=1:c
    for q = 1:l
        if matrix(q,w)==999999
            matrix(q,w) =NaN;
            nanidx = isnan(matrix(:,w));
            mean(matrix(~nanidx,w))
        end
    end
end

Grateful

[Walter Roberson]

You are not saving the result.

[Adam Danz]

It looks correct (I edited your comment and formatted your code with correct indentation but I didn't make any other changes). However, there are some other issues.

1) you're not storing the value of this function anywhere: mean(matrix(~nanidx,w))

2)If you're trying to replace all values of 999999 with NaN, you can do that more efficiently outside of the loop with this: matrix(matrix==999999) = NaN

3) Is this really what you want to do? You want to take the mean of column 'w' only if matrix(q,w) is 999999? That doesn't seem right.

[Guilherme Lopes de Campos]

The third option Adam,

I would to like replace the (999999) to mean of column (w);

Grateful

[Adam Danz]

If you want the mean of each column, ignoring 999999 and NaN values, this is one way you can do that:

matrix(matrix==999999) = NaN; 
nonNanCount = sum(~isnan(matrix),1);  %number of non-NaN per column
matrixTemp = matrix;                  %Temporarily replace matrix
matrixTemp(isnan(matrixTemp)) = 0;    %Replace NaN with 0
colMean = sum(matrixTemp,1)./nonNanCount; %mean of each column, ignoringNaN

[Adam Danz]

If you want the mean of each column only if there is a 999999 within the column and ignore that value, this is one way to do that:

% PRIOR to to the 5 lines in my comment above,
colIdx = any(matrix==999999,1); % Which columns have 999999 

% Then do the 5 lines in the comment above

% Then pull out the means from columns with 999999
colMean(colIdx)

[Guilherme Lopes de Campos]

Adam, thank you very much for help,

I need to replace the value where there is (999999) for a mean of

respective column,

I believe that is a little modification on code.

Grateful

[Adam Danz]

No problem; One of my comments above does that. It takes the mean of each column and ignores any 999999 values.

[Guilherme Lopes de Campos]

Hi Adam,

I tried the solution above, but it show the follow error:

Undefined function 'colMean' for input arguments of type 'logical'.

Error in depuracao_versao_2012 (line 13)
colMean(colIdx)

Could help me?

Grateful

[Adam Danz]

I've updated my answer to show a full working example with fake data.

[Guilherme Lopes de Campos]

Perfectly Adam,

Works,

But I need a help more,

How I can replace each value of colMean in the respectly column?

For example, I tried the code:

matrixTemp = matriz_media .* ~isKey; 
colMean = sum(matrixTemp)./rowCount;
colMean=transpose(colMean);

matrixTemp(matrixTemp==0)=colMean(1) 

This code replace all zero in matrix with the mean of position 1 (colMean(n,1)),but I need the mean of column 4 replace zero in column 4, mean of column 9 replace zero in column 9 like that respectly, I could use for looping?

I tried the follow code:

for i=1:c
    for j=1:l
        if matrixTemp(j,i)== 0
            matrixTemp(j,i)== colMean(i,1);
        end
    end
end

But not works

Could help me?

Very Grateful your patience.

Guilherme

[Adam Danz]

Happy to help.

To address your new question, try this out

[rowIdx,colIdx] = find(matrixTemp==0); 
matrixTemp(sub2ind(size(matrixTemp),rowIdx,colIdx)) = colMean(colIdx); 

[Guilherme Lopes de Campos]

Thank so much Adam,

I don't have words to show my thanks,

I am very grateful,

Guilherme

[Adam Danz]

I've been there; glad I could help!