Matchmaking

v1 = [ 5 -3 1 0 4 3 0 -5 4 -1]; v2 = [ 1 3 5 3 -4 -1 5 5 -1 4]; v3 = [-5 -2 3 -5 -4 -3 -3 1 -3 -3];Then by computing all the inner products (using

dot(v1,v2) = -51 dot(v1,v3) = -55 dot(v2,v3) = -11From this we might conclude that person 1 wouldn't really get along with either #2 or #3 but might get along slightly better with #2. #2 would be somewhat neutral with #3 (and vice-versa) and of the three #2 and #3 are the most compatible.

V = [v1',v2',v3']; C = V'*V 102 -51 -55 -51 128 -11 -55 -11 116For further processing (i.e. finding maximums) we need to first replace the diagonal elements with large negative values (as we know each person is compatible with themselves or at least we hope so, and we don't want your match to be you). Then we can find the maximum of each column. In MATLAB we do this, like:

D = diag(C); % save the diagonal elements C = C - diag(D) - 1000*eye(3); % replace the diagonal values with -1000 [z,ind] = max(C); % compute the max and its location of each columnThen z will contain the maximum of each column, while ind(i) will be the number of the person most compatible with person i. In our sample problem, you'd get z = (-51,-11,-11) and ind = (2,3,2). Meaning person 1 is most compatible with person 2, and persons 2 and 3 are most compatible with each other.

Once you have the answers, process them to find the most compatible person for each person.

Turn in your questionaire and the results (you can just give me the numbers, i.e. the ind vector from above).

2. (Extra Fun) In a more scientific situation the approach would be similar except
that the questions would be chosen more carefully and instead of
using the standard Euclidean norm, they would use some sort of
weighted norm. For example if they thought that similarities or
differences in one category had more impact, they would weight
the inner product in favor of that question, i.e. instead of
using the term u_{i}v_{i} to contribute to
the inner product, they'd use 10u_{i}v_{i}.
Also, because answers to questions might be linked they might
introduce a matrix A representing those links, e.g. instead
of using the answers to 1 and 2 separately they might look at
a combination of those two answers before they did the inner product.
In this case, you'd compute the compatibility matrix via V'*A'*A*V.

Create a 10 x 10 matrix A which changes the weights on each question or combines the answers to the questions and compute the compatiblilty results for your sample.

Turn in the matrix and the results.