cumsum within a group

Hi, Is there a faster way to do a cumulative sum within a group? data looks like

    year val cumval
data=[1  10  10;
1  10  20;
1  10  30;
1  10  40;
2  10  10;
2  15  25;
2  20  45;
3  5  5;
4  10  10];

As you see cumVal is sumsum within a year only. Then what I want to do is to compute a new val that reflects this yearly cap. I create 2 more columns. cumval_w_annualCap reflects the annual cap value of 25 in this example. The val_cap is the re-compute val that reflects

cumval_w_annualCap val_cap
tcumsum=[10  10;
20  10;
25  5;
25  0;
10  10;
25  15;
25  0;
5  5;
10  10];

How I compute it seems to be very slow.

val_cap= data( :,2);

yearlyLim=25;
[ uniqYrs, indxYr] = unique( data(:,1));
nUniqYr = length(uniqYrs);
for runYr= 1: nUniqYr
    if runYr< nUniqYr
        tindx= [indxYr( runYr):indxYr( runYr+1)-1];
    else
        tindx= [indxYr( runYr):length( data(:,2))];
    end % 
    yrVal   = data(tindx,2);
    tcumval= cumsum( yrVal, 1);
    indxAggCap = find(tcumval >= yearlyLim, 1);

      if indxAggCap == 1
          yrVal( indxAggCap)= yearlyLim;
      else
          yrVal( indxAggCap)= yearlyLim - tcumval( indxAggCap-1);
      end % if indxAggCap
      yrVal( indxAggCap+1:end)= 0;
      val_cap( tindx)= yrVal;

end %

Anyway to vectorize this algorithm?

Thanks,

Answer 1

[Thorsten]

 data=[1  10  10;
1  10  20;
1  10  30;
1  10  40;
2  10  10;
2  15  25;
2  20  45;
3  5  5;
4  10  10];
  % first column of tcumsum
 yearlyLim=25;
 t1 = data(:,3);
 t1(t1 > yearlyLim) = yearlyLim;

 % second column of tcumsum
 [~, b] = unique(data(:,1));
 Nb = numel(b);
 t2 = arrayfun(@(i)  [t1(b(i)); diff(t1(b(i):b(min(i+1, Nb))-1))]', 1:Nb, 'Uni', 0);
 t2 = cell2mat(t2)';
 tcumsum = [t1 t2];

Since R2015b you can use split apply

t2 = cell2mat(splitapply(@(x) {[x(1); diff(x)]}, t1, data(:,1)));

Answer 2

[Andrei Bobrov]

data=[1  10 
1  10 
1  10 
1  10 
2  10 
2  15 
2  20 
3  5 
4  10];

[~,~,c] = unique(data(:,1));
d = data(:,2);
o1 = cell2mat(accumarray(c,d,[],@(x){cumsum(x)}))
val_cap = cell2mat(accumarray(c,min(o1,25),[],@(x){[x(1);diff(x)]}));