Reducing the time is takes to calculate a large 3D matrix?

Question

Conal Doyle 2013-7-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/82775-reducing-the-time-is-takes-to-calculate-a-large-3d-matrix

I'm trying to populate a 3d matrix with profit and loss values for cells defined ccs, cp and woy which are unique entries of my original data currencyPair, counterpartyAlias and weekofYear cells respectively. Unfortunately, length of ccs is 117, cp is 1380 and woy is 25 so the matrix takes a couple of days to calculate. However, the majority of rows and columns are zeros because most counterparties only deal across selective currency pairs. Is there a way to populate the matrix without calculating the zero entries? I tried concatenating the ccs and cp cells before looking for the unique entries but then I have nothing to use a strcmp against. Any ideas would be greatly appreciated!

PnL_CCSvsCP = zeros(length(ccs),length(cp),length(woy));
for i=1:length(ccs)
  for j=1:length(cp)
    for k=1:length(woy)
      PnL_CCSvsCP(i,j,k) = sum(PnL(strcmp(currencyPair,ccs{i}) & ...
                               strcmp(counterpartyAlias,cp{j}) & ...
                               (weekofYear == woy(k))));
    end
  end
end

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jan 2013-7-22

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/82775-reducing-the-time-is-takes-to-calculate-a-large-3d-matrix#answer_92412

编辑：Jan 2013-7-22

在 MATLAB Online 中打开

PnL_CCSvsCP = zeros(length(ccs), length(cp), length(woy));
for i=1:length(ccs)
  m1 = strcmp(currencyPair, ccs{i});
  for j=1:length(cp)
    m2 = m1 & strcmp(counterpartyAlias,cp{j});
    for k=1:length(woy)
      PnL_CCSvsCP(i,j,k) = sum(PnL(m2 & (weekofYear == woy(k))));
    end
  end
end

Resorting the loops might help also, such that the longer loops are moved to inside:

m_cp = false(numel(counterpartyAlias), numel(cp));  % Or transposed?
for j = 1:length(cp)
  m_cp(:, j) = strcmp(counterpartyAlias, cp{j});
end
PnL_CCSvsCP = zeros(length(ccs),length(cp),length(woy));
for k=1:length(woy)
  m_woy = (weekofYear == woy(k));
  for i=1:length(ccs)
    m_ccs_woy = strcmp(currencyPair, ccs{i}) & m_woy;
    for j = 1:length(cp)
      PnL_CCSvsCP(i,j,k) = sum(PnL(m_ccs_woy & m_cp(:, j));
    end
  end
end

And now you can try to replace the innermost loop by a BSFXUN call.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Jan 2013-7-22

编辑：Jan 2013-7-22

I do not understand your question. Of course the modified code does exactly what your code does. But I tried to reduce the number of repeated calculations. In the 2nd method e.g. the string comparison with cp is done 25 instead of 4036500 times as in your original version. This should reduce the runtime remarkably. Did you test this already?

Without having any test data, improvements are based on some guessing. Further improvements are possible, when you give us a hint, whether the current suggestions are working at all. Notice that I've guessed e.g. the orientation of the cell strings etc.

Conal Doyle 2013-7-23