Avoid loops to make code run faster

Hi Everyone,

I have some code I want to make faster, and therefore I want to delete my for loops. I have a matrix, TTF_Live_Q, with 3 columns – datenums, bid prices & ask prices. In that matrix I have quarterly prices, so there is only a date/price in every third cell. It could look like this:

0  0  0  

0  0  0

735690  26.90  27

0  0  0  

0  0  0  

735781  28.10  28.20

0  0  0  

0  0  0

I want the prices in 2. & 3. column to be repeated 3 times, this I have done with:

    for i=1:length(TTF_Live_Q)
        if L_Q(i) == 1
            TTF_Live_Q(i+1,2:3) = TTF_Live_Q(i,2:3);
            TTF_Live_Q(i+2,2:3) = TTF_Live_Q(i,2:3);
        end 
    end

Where L_Q is an index vector. The matrix will now look like:

0  0  0  

0  0  0

735690  26.90  27

0  26.90  27  

0  26.90  27  

735781  28.10  28.20

0  28.10  28.20  

0  28.10  28.20

I have tried kron, but that doesn’t work, as there are 0 between the quarterly prices. I am sure you guys have a smarter way to do it.

Next thing I want to improve is the loop for the dates. I need the date number for the beginning of each month. It should look like this:

0  0  0  

0  0  0

735690  26.90  27

735720  26.90  27  

735751  26.90  27  

735781  28.10  28.20

735812  28.10  28.20  

735843  28.10  28.20

I have a vector with all the correct dates; Here is the loop that makes the date numbers come into the matrix.

    Index_Q = zeros(size(TTF_Live_Q,1),1);  

    for i=1:3:size(TTF_Live_Q,1)-2
        Index_Q(i) = find(TTF_Live_Q(i,1)==Date_Vector);
        TTF_Live_Q(i+1,1) = Date_Vector(Index_Q(i)+1); 
        TTF_Live_Q(i+2,1) = Date_Vector(Index_Q(i)+2);
    end

Can you please tell me a smarter way to do it? Thanks a lot!

Answer 1

[Andrei Bobrov]

data = [0 0 0  
0 0 0
735690 26.90 27
0 0 0  
0 0 0  
735781 28.10 28.20
0 0 0  
0 0 0];

out = data;
l = all(cumsum(data)~=0,2);
d0 = data(l,:);
[y,m,dd] = datevec(d0(1));
out(l,1) = datenum(y,m+(0:size(d0,1)-1)',dd);

d1 = d0(:,2:3);
l1 = d1(:,1)>0;
d2 = d1(l1,:);
out(l,2:3) = d2(cumsum(l1),:);

ADD

data = [0 0 0  
0 0 0
735690 26.90 nan
0 0 0  
0 0 0  
735781 28.10 28.20
0 0 0  
0 0 0];

out = data;
l = all(cumsum(data)~=0,2);
d0 = data(l,:);
[y,m,dd] = datevec(d0(1));
out(l,1) = datenum(y,m+(0:size(d0,1)-1)',dd);

d1 = d0(:,2:3);
p = d1(all(d1>0 | isnan(d1),2),:);
ii = find(isnan(p));
p(ii) = p(ii+1);
out(l,2:3) = kron(p,ones(3,1));

[Betina Isbak]

Hi Andrei,

It works like a charm. However, if there isn't a price for a quarter then the cell will have NaN in it. If the cell has NaN in it, you way will replace the NaN value with the price from the last quarter. I want it to copy the NaN values down like a normal price. Is that possible?

Thank you very much for your help!

[Simon]

Hi!

Do you always want to copy the row (3+n*3) to the two rows below? Then you can write

num = size(data, 1);
copyrows = 3:3:num;
data(copyrows+1, :) = data(copyrows, :);
data(copyrows+2, :) = data(copyrows, :);

[Betina Isbak]

Hi Simon,

That works great for the prices, but not for the dates. The dates has to change for every row, so I get the correct month for the price.

Thanks!

[Andrei Bobrov]

Hi Betina! See ADD part of my answer.

[Simon]

So, you can mix it with Andrei's answer. Just replace in my answer the ":" with "2:3" in the last two rows.

[Betina Isbak]

Yeah, that was my thought too.. It works great! Thank you to both of you..