extract user defined data from matrix in GUI

This might be a simple question but I still need some help to sort this out. I have Dataset and from this the user defines which columns to extract from a uitable. I can get the uitable data which will look like the "sample" below:

Dataset =

     1     2     3     4     5     6
     5     6     7     8     9    10
    10    20    30    40    50    60
   100   200   300   400   500   600

 sample =

     1     3     0
     2     4     6

Now I want to extract the columns specified in each "sample" row of the Dataset and calculate the mean and finally create a new matrix with then calculated means in each column. The result would be:

new =
     2     4
     6     8
    20    40
   200   400

Sorry for the clumsy description, but I hope I makes sense. But any help is very appreciated. The above example is a simplified example. The Dataset I ahve is very large andso is the "sample" set (many rows x 3 columns),so I believe a loop will solve the issue but please let me know if you have other ideas. Thanks

[Jan]

I read the question 5 times now, without understanding it. You want to extract columns from matrix DataSet. These columns are "specified in each row of sample". The first row is [1,3,0]. Perhaps this means the 1st and 3rd column of DataSet. But then the mean would be [2, 6, 20, 200]. Is the leading 1 a typo?

Now the 2nd row: [2,4,6]. The corresponding columns are:

       2     4     6
       6     8    10
      20    40    60
     200   400   600

And again the mean is not [4, 12, 40, 400], but the 2nd element would be 8.

Please clarify this.

[JB]

Hi Jan Simon. You are absolutely right. I have corrected in the question above. Thanks a lot.

Answer 1

[Jan]

DataSet = [ ...
   1     2     3     4     5     6; ...
   5     6     7     8     9    10; ...
  10    20    30    40    50    60; ...
 100   200   300   400   500   600]; 
sample = [1     3     0; ...
          2     4     6];
nSample = size(sample, 1);
Avg     = zeros(size(DataSet, 1), nSample);
for k = 1:nSample
  index     = sample(k, :);
  index     = index(index ~= 0);
  Avg(:, k) = sum(DataSet(:, index), 2) / numel(index);  % Faster than mean()
end

Does this produce the wanted output? If so, let's think of vectorizing this:

wData              = size(DataSet, 1);
[nSample, wSample] = size(sample);
nValidSample       = sum(sample ~= 0, 2).';

% Insert a column of zeros to handle the 0 in sample:
DataPad = cat(2, zeros(wData, 1), DataSet);

% Shift the indices in sample and use them as column index:
Sampled = DataPad(:, sample.' + 1);

% Create blocks to sum over:
Sampled = reshape(Sampled, wData, wSample, nSample);

% Average over blocks:  (! Auto-Expand, >= R2016b !)
Avg = reshape(sum(Sampled, 2), wData, nSample) ./ nValidSample;

For older Matlab versions the last line is slightly slower:

Summed = reshape(sum(Sampled, 2), wData, nSample);
Avg    = bsxfun(@rdivide, Summed, nValidSample);

A speed comparison (R2016b/64):

DataSet = rand(4, 1e3);
sample  = randi([0, 1e3], 1e6, 3);

 Loop:       6.46 sec
 Vectorized: 0.36 sec

[JB]

BRILLIANT. It does exactly what I want. Thanks a million.

[Jan]

Copied from the section for answers: [JB wrote:]

@Jan Simon. Once again thanks a lot for your great help. I am working with MATLAB R2015b/32 and get an error in the last code line:

Error using ./ Matrix dimensions must agree.

Am i making a mistake or???

[Jan]

Please read my answer carefully. I repeat:

 % Average over blocks:  (! Auto-Expand, *** >= R2016b *** !)
 Avg = reshape(sum(Sampled, 2), wData, nSample) ./ nValidSample;

For older Matlab versions the last line is slightly slower:

 Summed = reshape(sum(Sampled, 2), wData, nSample);
 Avg    = bsxfun(@rdivide, Summed, nValidSample);