I have a collect of N vectors, of arbitrary dimension,k. So they can be arranged as a k column by N row matrix. I want to sort them by significance, the vectors which most precisely describe the collect sorted higher.
Step one, find the mean/variance of each column. The column with the highest m/variance is the more significant, it has the lease noise. I pick some base b such that the logb(1+m^2/var) for all columns that makes the most significant column have a log at least one great then the rest. I take the log of each element in the most significant column.I sort the rows based on that column. I split the matrix in two groups depending on whether the most significant bit is set in that column.
I then repeat this process for each of the two resulting matrices until there is no more sorting to do. Does this work?
No comments:
Post a Comment