How to sum specific values from a loop and create a new vector?

If I have this matrix M:

 a=1:1:20;
 b=[0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 2 0 0];

 M=[a,b];

How could I find the i values where M(:,b)>0 (where the second column "b">0), and then take the values of the same row but the first column "a" M(a,:) and make a summation from that row i and i-2, i,e, [i + (i-1)+(i-2)].. like this:

 X=[(5+6+7) (12+13+14) (18+17+16)] = [18 39 51]

Besides a vector Y which contains the summation from i-5 until i-3 like this:

Y=[(2+3+4) (9+10+11) (13+14+15)] = [9 30 42]

Both of them for all i values in a vector b of a matrix.

Answer 1

[Star Strider]

Use bsxfun and sum:

a=1:1:20;
b=[0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 2 0 0];
M=[a(:),b(:)];
idx = find(M(:,2) > 0);
X = sum(bsxfun(@plus, idx, -2:0),2)
Y = sum(bsxfun(@plus, idx, -5:-3),2)

X =
    18
    39
    51

Y =
     9
    30
    42

[Philippe Corner]

Thank you man!, one question: X = sum(bsxfun(@plus, idx, -2:0), 2)

what does this 2 implies?

[Star Strider]

My pleasure!

The 2 is an argument to sum, and tells it to sum across columns. (The default behaviour is to sum across rows, or dimension 1.)

[Philippe Corner]

Mr Star, im not sure what is happening, I modified the code in function of what im needing and the result is not coherent.. can you take a look? Im trying to sum 10 rain values before each event>0

clear all
[date,rain,event] = textread('date_rain_event.txt','%s %f %f','delimiter',',', 'headerlines',1);

M=[rain,event] ;

idx = find(M(:,2) > 0);
f = sum(bsxfun(@plus, idx, -9:0),2)

[Star Strider]

Originally, ‘idx’ and ‘M(:,1)’ were the same. The code using ‘M(:,1)’ is:

X = sum(bsxfun(@plus, M(idx,1), -2:0),2)
Y = sum(bsxfun(@plus, M(idx,1), -5:-3),2)

giving the same result.

So this should work in your code:

f = sum(bsxfun(@plus, M(idx,1), -9:0),2)

My apologies for the confusion.

[Philippe Corner]

Thank you again, you are very kind. I got your point.

Nevertheless, it is not working.. idx is doing good, the first row where M(:,2)>0 is 127.. and it should be taking the rowsum of the first column like this: (118:127, 1), but the result is not ok, and even there are negative values now..

Ill keep working on it, but if you can easily find the mistake, please let me know.

Thank you again.

[Star Strider]

My pleasure.

I still do not understand the reason that the prototype code did not work with your actual data.

This works:

f = sum(arrayfun(@(x)sum(M(x,1)), bsxfun(@plus, idx', (-9:0)')));

I checked to be certain it produces the correct result with your data.

It works by calculating one column vector of indices into ‘M’ for each value of ‘idx’ and the offset of ‘(-9:0)’, calculated by the bsxfun call (that turned out to be correct after all), creating a matrix that here is (10x1881), the column size being the length of ‘idx’. It then uses the custom anonymous function ‘@(x)sum(M(x,1))’ to sum the first column of ‘M’ corresponding to the indices calculated by the bsxfun call. The arrayfun function contains an implicit loop (that I hope is more efficient than an explicit for loop). It then returns the result of the row sums for each column as a row vector.

[Philippe Corner]

Amazing! The last question would be how to solve the same problem but not for the rows which M(:,2)>0, but all the rows! if I apply the same sense, the problem would be for the very beginning rows.. any idea Mr Star?

[Star Strider]

Thank you!

If you want to have a moving sum for all the rows taken 10 at a time, use the movsum (link) function (introduced in R2016a).