Post by Doc_Lobster on Sept 28, 2012 20:31:21 GMT -5
I had to write a program that can simulate randomly distributed clusters of data points with minimal user inputs. I was kind of stuck on what to use for the covariance matrices (because I don't want them to be the same for each cluster, too simplistic). Turns out that the eigenvalues of the covariance matrix are the squares of the standard deviation of the distance from the centroid in each dimension of the cluster. If I use a distribution of standard deviations and a generic set of eigenvectors, I can make several unique covariance matrices. Walla!
Then I had one of the grad students in my office trying to get me to "just" include a simulation of blah blah blah (super ass complicated shit) because that would be great. I hate it when people say "can't you just do this thing?" without any conception of how difficult it would actually be.
I had to write a program that can simulate randomly distributed clusters of data points with minimal user inputs. I was kind of stuck on what to use for the covariance matrices (because I don't want them to be the same for each cluster, too simplistic). Turns out that the eigenvalues of the covariance matrix are the squares of the standard deviation of the distance from the centroid in each dimension of the cluster. If I use a distribution of standard deviations and a generic set of eigenvectors, I can make several unique covariance matrices. Walla!
Then I had one of the grad students in my office trying to get me to "just" include a simulation of blah blah blah (super ass complicated shit) because that would be great. I hate it when people say "can't you just do this thing?" without any conception of how difficult it would actually be.