- #1
kmrstats
- 2
- 0
Hi. I have a problem I am hoping you all can shed some light on.
I have N entities, O, each described by N values - a weight W and N-1 similarity coefficients to the other N-1 entities. I guess we can represent Oi as (Wi, Sij, j=(1,2,...,N, i!=j)(?).
Given an integer M and M < N I need to maximize the following function:
Sumi(Wi/Sumj(Sij, j=(1,2,...,N), i!=j))
where i and j in the sums are constrained to the unique combinations of the integers 1 to N of size M.
As an example let N=4, M=3 there are then 4 unique combinations of the numbers 1,2,3,4 of size 3: (1,2,3), (1,2,4), (1,3,4) and (2,3,4). The values for the function are therefore:
W1/(S1,2+S1,3)+W2/(S2,1+S2,3)+W3/(S3,1+S3,2)
W1/(S1,2+S1,4)+W2/(S2,1+S2,4)+W4/(S4,1+S4,2)
W1/(S1,3+S1,4)+W3/(S3,1+S3,4)+W4/(S4,1+S4,3)
W2/(S2,3+S2,4)+W3/(S3,2+S3,4)+W4/(S4,2+S4,3).
For small N and M I can enumerate the combinations and calculate the maximum sum but my problem has N~50 and M~5.
Any suggestions or thoughts of an akgorithm to calulate the max? I hope I have made it clear.
Thanks in advance.
I have N entities, O, each described by N values - a weight W and N-1 similarity coefficients to the other N-1 entities. I guess we can represent Oi as (Wi, Sij, j=(1,2,...,N, i!=j)(?).
Given an integer M and M < N I need to maximize the following function:
Sumi(Wi/Sumj(Sij, j=(1,2,...,N), i!=j))
where i and j in the sums are constrained to the unique combinations of the integers 1 to N of size M.
As an example let N=4, M=3 there are then 4 unique combinations of the numbers 1,2,3,4 of size 3: (1,2,3), (1,2,4), (1,3,4) and (2,3,4). The values for the function are therefore:
W1/(S1,2+S1,3)+W2/(S2,1+S2,3)+W3/(S3,1+S3,2)
W1/(S1,2+S1,4)+W2/(S2,1+S2,4)+W4/(S4,1+S4,2)
W1/(S1,3+S1,4)+W3/(S3,1+S3,4)+W4/(S4,1+S4,3)
W2/(S2,3+S2,4)+W3/(S3,2+S3,4)+W4/(S4,2+S4,3).
For small N and M I can enumerate the combinations and calculate the maximum sum but my problem has N~50 and M~5.
Any suggestions or thoughts of an akgorithm to calulate the max? I hope I have made it clear.
Thanks in advance.