Table manipulation of matrix

Hi

I want to a code that will compute group each c1/c2 for each column using K = to any number

Assuming I have n = numbers length

In my case I have 1,200 columns

And 1,200 rows

For instance, if K = 2

For each columns

The sum lowest c1 + c1 / the sum lowest c2 + c2

And it goes down the column for the next lowest c1 + c1/c2 + c2 till the end

If K = 3

Same the lowest c1 + c1 + c1/ the lowest c2 + c2 + c2

And it goes down the column for the next lowest c1 + c1 + c1/c2 + c2 + c2 till the end

I have been trying to work this out using the code below but but not getting it right. It is supposed to give me multiple answers for each column depending on K

c1 = sort(J(1:2:end,:));
size(c1)
c2 = sort(J(2:2:end,:));
size(c2
f = sum (c1(1: K, :));   
size(f)
c = sum (c2(1: K,: ));
size(c)
s1 = (f/c);

for instance find the data below

A    B    C    D   E  F   G  Class
1    1    1    0   1  1   1    c1
2    2    1    2   1  0   1    c2
1    1    1    1   2  1   1    c1
2   1     1     1   1  1   1   c2
2   2     0     0   0  0   0   c1
1   1     1     1   1  1   1   c2
1   1     1     1   1  1   1   c1
1   0     0     1   1  1   1   c2

Lets do column 1

First Sorting

1    c1
1    c1
1    c1
1    c2
1    c2
2    c1
2    c2
2    c2

K = 2

The lowest c1 + c1/ The lowest c2 + c2

For column A

1+1/1+1 = 2/2 = 1

The next lowest c1 + c1/The next lowest c2 + c2

1 + 2 / 2+2 = 3/2 = 1.5

So column 1 will be

1
1.5

Thanks for your help in advance

Tino

Find my code again

c1 = sort(J(1:2:end,:));
size(c1)
c2 = sort(J(2:2:end,:));
size(c2
f = sum (c1(1: K, :));   
size(f)
c = sum (c2(1: K,: ));
size(c)
s1 = (f/c);

Answer 1

[Guillaume]

Your question is really badly explained but I think I've understood what you want.

My understanding is that you have a J matrix where the odd rows are labeled 1 and even rows labeled 2. For each column, you want to sum the sorted odd rows over a non-overlapping sliding window of length K and divide that by the corresponding sum over the sorted even rows.

If so:

%J: a matrix whose number of rows is a multiple of 2*K. 
%K: an integer, the window size
assert(mod(size(J, 1) / 2, K) == 0, 'Height of J is not a multiple of 2*K')

result = zeros(size(J, 1) / 2 / K, size(J, 2));  %preallocate result
for col = 1:size(J, 2)  %loop over the columns. You can't do this without a loop due to the requirement to sort each columns separately
    result(:, col) = sum(reshape(sort(J(1:2:end, col)), K, []), 1) ./ sum(reshape(sort(J(2:2:end, col)), K, []), 1);  %sort even/odd rows, reshape each into a Kx? matrix, sum across rows and divide the two sums
end

[Tino]

Hi Guillaume

Thanks for your response.

I run the code and am getting

Undefined function or variable 'L'.

What does L means

Thanks in advance

Tino

[Guillaume]

Basic typo, that L should have been K. Letters are next to each other on the keyboard.

Fixed now.

[Tino]

Hi Guillaume

I am experimenting the code you gave to me on a sample matrix

and I am getting the error below

Error using fishy (line 17)

Height of J is not a multiple of 2*K

why?

find the sample dataset below

T =[

5.1000 3.5000 1.4000 0.2000;

6.4000 3.2000 4.5000 1.5000;

5.4000 3.4000 1.7000 0.2000;

5.7000 2.8000 4.5000 1.3000;

5.7000 4.4000 1.5000 0.4000;

5.6000 2.9000 3.6000 1.3000];

z = T(:,:);

K = 2;

assert(mod(size(z, 1) / 2, K) == 0, 'Height of J is not a multiple of 2*K')

result = zeros(size(z, 1) / 2 / K, size(z, 2)); %preallocate result

for col = 1:size(z, 2)

%loop over the columns. You can't do this without a loop due to the requirement to sort each columns separately

result(:, col) = sum(reshape(sort(z(1:2:end, col)), K, []), 1) ./ sum(reshape(sort(z(2:2:end, col)), K, []), 1); %sort even/odd rows, reshape each into a Kx? matrix, sum across rows and divide the two sums

end

size(result)

Error using fishy (line 17)

Height of J is not a multiple of 2*K

[Guillaume]

As I wrote in a comment on the first line of the code, the number of rows of J must be a multiple of 2*K. My code then check that it is the case and if not, throws the error you see.

Since you're dividing the odd (and even) rows in groups of length K, the number of odd (and even) rows must be a multiple of K, otherwise one of the window can't be of length K. And since you need the same number of odd and even rows, the height of J must be a multiple of 2*K.

If you don't want the code to error if that precondition is broken, you need to explain what should be done then.

[Tino]

Hi Guillaume

Thanks for your response

But a given data might not really satisfy the condition 2K. in that case the calculation must stop and the answers provided from before the condition is not met instead of the error.

And how do I do that

Regards

Tino

[Guillaume]

So basically, you want to crop J to a multiple of 2*K:

%J: a matrix of any size
%K: window size
croppedJ = J(1:end-mod(end, 2*K), :);  %ensure that height of J is a multiple of 2*K by removing extra elements from the end
result = zeros(size(croppedJ, 1) / 2 / K, size(croppedJ, 2));  %preallocate result
for col = 1:size(croppedJ, 2)  %loop over the columns. You can't do this without a loop due to the requirement to sort each columns separately
    result(:, col) = sum(reshape(sort(croppedJ(1:2:end, col)), K, []), 1) ./ sum(reshape(sort(croppedJ(2:2:end, col)), K, []), 1);  %sort even/odd rows, reshape each into a Kx? matrix, sum across rows and divide the two sums
end

[Tino]

You are brilliant Guillaume and Thanks a lot

Assuming I have 3 classes and wishes to use this code above. The classes ( for instance c1, c2, c3) can assigned to any regardless of if it is even or odd.

Can I just replace where u have 2 with 3 or will it work that way

Thanks in advance

[Guillaume]

If you had more than two classes, I would have used a completely different method to identify the elements of each. It still would be very simple but nothing like the code above.

However, I don't see how you can extend the code to more than two classes, since the main step is dividing the elements of one class by the elements of the others. If you have more than 2 classes, what are you dividing by what?

[Tino]

Thanks for the answer. What if I have 3 classes what is the completely different method you will use to identify it.

Thanks for your persistence help

I really appreciate

Regards

Tino

[Guillaume]

You'd use findgroups or the older unique to identify the groups, then splitapply or the older accumarray to apply the processing to each group.