A time killing loop

Hi,

I have a loop which is taking ages. I am wondering if someone can make it more efficient.

%%x1 is a vector of date rows for which I need to find the price
 for k=1:length(x1)

%%x2 finds the rows in the original data file for each specific date (A is a vector of 6320 dates,ID is the unique identifier of a bond issued by a firm. The first column in data is the column of unique bond identifiers and the second column is a column of dates . Finally,the third column of data is the column of prices )

x2=find(data(:,2)==A(x1(k))&data(:,1)==ID);

%%%IssueMatrix is a matrix of prices; the first column is equal to the date vector A , and then each column refers to the prices of a different bond j issued by a firm.

 if isnan(IssueMatrix(x1(k),j+1))==1

%%if I don't have any price on a date A(x1(k) for a specific bond issue then assign a price which is the average price of bonds supplied by different bond dealers.
     IssueMatrix(x1(k),j+1)=nanmean(data(x2,3));
        else

%%%if I have a price assigned for this bond added it to other prices supplied by dealers and then take the average         

IssueMatrix(x1(k),j+1)=nanmean([data(x2,3);IssueMatrix(x1(k),j+1)]);   
          end

[Jan]

Please format the posted code and provide some test data, such that we can read and run your code.

[joseph Frank]

I did my best to simplify the codes and to post some sample data. I hope these can be understood easily.

Answer 1

[Jan]

Did you pre-allocate IssueMatrix ?

Can you create the IssueMatrix at first and apply FEX:nancumsum afterwards?

[joseph Frank]

as per the codes in my second post I created IssueMatrix (IM in the new codes) when I applied 0 votes

Jan Simon answered about 2 hours ago

Did you pre-allocate IssueMatrix ?

Can you create the IssueMatrix at first and apply FEX:nancumcum I received an error.

Answer 2

[joseph Frank]

I changed the names of the variables so that they can appear in a readable format and I cut my huge data to a very small sample to make it available for you.Kindly, can we speed up the loop ?(it is fast for the data posted because of the small data size but for my large data it is extremely slow.

    UIr=unique(I(:,2));
    for i=1:length(UIr)

        x0=find(I(:,2)==UIr(i)); 
        UI=unique(I(x0,1));
        IM=nan(length(A),length(UI)+1);
        IM(:,1)=A;

        for j=1:size(UI,1)
            for j=1:size(UI,1)
            ID=UI(j,1);
            ID2=repmat(ID,size(A));
            x1=find(ismember(IM(:,1),unique(D(:,2))) & ismember(ID2,D(:,1)));
            for k=1:length(x1)
            x2=find(D(:,2)==A(x1(k))& D(:,1)==ID);
            if isnan(IM(x1(k),j+1))==1
            IM(x1(k),j+1)=nanmean(D(x2,3));
            else
           IM(x1(k),j+1)=nanmean([D(x2,3);IM(x1(k),j+1)]);   
            end
            end
            end
       end
     end

   A=728185
  728186
  728187
  728188
  728189
  728190
  728191
  728192
  728193
  728194
  728195
  728196
  728197
  728187
  728188
  728189
  728190
  728191
  728192
  728193
  728194
  728195
  728196
  728197
  728198
  728199
  728200
  728201];

D=[458  731015  100  133
458  731015  100  133
458  731015  100  133
458  730948  100  133
458  730463  103  133
458  730463  103  133
458  729837  108  133
458  729722  110  133
458  729574  111  133
458  728187  111  133
475  728188  111  133
475  728189  111  133
475  728190  111  133
475  728191  111  133
475  728192  111  133
475  728193  111  133
475  728194  111  133
475  728195  111  133
475  728196  111  133
475  728197  110  133
475  728198  116  133
475  728199  117  133
475  728200  116  133
475  728201  115  133
475  728187  113  133
475  728188  110  133
475  728189  108  133
475  728190  109  133
475  728191  105  133
475  728192  105  133
475  728193  109  133
475  728194  113  133
475  728195  105  133
475  728196  108  133
475  728197  110  133
475  728198  110  133
475  728199  109  133
475  728200  109  133
475  728201  107  133
475  728187  107  133
475  728188  111  133
475  728189  110  133
475  728190  108  133
475  728191  107  133
456  728192  108  133
456  728193  107  133
456  728194  108  133
456  728195  104  133
456  728196  100  133
456  728197  100  133
456  728198  113  133
456  728199  114  133
456  728200  113  133
456  728201  100  133
456  728187  113  133
456  728188  116  133
456  728189  100  133
456  728190  116  133
456  728191  111  133
456  728192  112  133
456  728193  111  133
456  728194  112  133
456  728195  103  133
456  728196  103  133
456  728197  103  133
456  728198  103  133
456  728193  103  133
456  728194  104  133
456  728195  104  133
456  728196  104  133
456  728193  104  133
456  728194  104  133
456  728195  104  133
456  728196  107  133];

I=[456  133  3353
458  133  3353
475  133  1021];

[Jan]

You have two nested loops using j as counter. I assume this is a bug.

[joseph Frank]

oh . Thanks a million Jan

[Jan]

Replace:

ID=UI(j,1); ID2=repmat(ID,size(A));

x1=find(ismember(IM(:,1),unique(D(:,2))) & ismember(ID2,D(:,1)));

by:

x1 = find(ismember(IM(:,1),unique(D(:,2))) & (UI(j)==D(:,1)));

Answer 3

[Oleg Komarov]

This a vectorized way to obtain intersected means, then it's up to you to replicate the matrix anc concatenate as you wish:

% Row subs

[idx,rsub] = ismember(D(:,2),A);
Dmemb = D(idx,:);
rsub   = rsub(idx);

% Col subs

[idx,csub] = ismember(Dmemb(:,1),I(:,1));

% Accumarray

out = [unique(Dmemb(:,2)) accumarray([rsub-min(rsub)+1, csub],Dmemb(:,3),[],@mean,NaN)]