Question

Given a list of values (e.g. 10, 15, 20, 30, 70), values N (e.g. 3) and S (e.g. 100), find a subset that satisfies :

1. length of subset >= N
2. sum of subset >= S

The sum of the subset should also be the least possible (the sum of remaining values should be the greatest possible) (e.g. result subset should be (10,20,70), not (15,20,70) which also satisfies 1. and 2.).

I was looking at some problems and solutions (Knapsack problem, Bin packing problem, …) but didn’t find them applicable. Similar problems on the internet were also not suitable for some reason (e.g. number of elements in subset was fixed).

Can someone point me in the right direction? Is there any other solution other than exhausting every possible combination?

Edit – working algorithm I implemented in ruby code, I guess it can be further optimized:

def find_subset_with_sum_and_length_threshold(vals, min_nr, min_sum)
    sum_map = {}
    vals.sort.each do |v|
      sum_map.keys.sort.each do |k|
        addends = sum_map[k] + [v]
        if (addends.length >= min_nr && k+v >= min_sum)
          return addends
        else
          sum_map[k+v] = addends
        end
      end
      sum_map[v] = [v] if sum_map[v].nil?
    end
  end

Answer 1

This is not very different from the 0-1 knapsack problem.

Zero-initialize a matrix with S+U rows and N columns(U is the largest list value)
Zero-initialize a bit array A with S+U elements
For each value (v) in the list:
  For each j<S:
    If M[N-1,j] != 0 and M[N-1, j + v] == 0:
      M[N-1, j + v] = v
      A[j + v] = true
  For i=N-2 .. 0:
    For each j<S:
      If M[i,j] != 0 and M[i+1, j + v] == 0:
        M[i+1, j + v] = v
  M[0,v] = v
Find first nonzero element in M[N-1,S..S+U]
Reconstruct other elements of the subset by subtracting found value from its\
  index and using the result as index in preceding column of the matrix\
  (or in the last column, depending on the corresponding bit in 'A').

Time complexity is O(L*N*S), where L is the length of the list, N and S are given limits.

Space complexity is O(L*N).

---

Zero-initialize an integer array A with S+U elements
i=0
For each value (v) in the list:
  For each j<S:
    If A[j] != 0 and A[j + v] < A[j] + 1:
      A[j + v] = A[j] + 1
      V[i,j + v] = v
      P[i,j + v] = I[j]
      I[j + v] = i
  If A[v] == 0:
    A[v] = 1
    I[v] = i
  ++i
Find first element in A[S..S+U] with value not less than N
Reconstruct elements of the subset using matrices V and P.

Time complexity is O(L*S), where L is the length of the list, S is given limit.

Space complexity is O(L*S).

---

Algorithm that also minimizes the subset size:

Zero-initialize a boolean matrix with S+U rows and N columns\
  (U is the largest list value)
Zero-initialize an integer array A with S+U elements
i=0
For each value (v) in the list:
  For each j<S:
    If A[j] != 0 and (A[j + v] == 0) || (A[j + v] > A[j] + 1)):
      A[j + v] = A[j] + 1
      V[i,N-1,j + v] = v
      P[i,N-1,j + v] = (I[j,N-1],N-1)
      I[j+v,N-1] = i
  For k=N-2 .. 0:
    For each j<S:
      If M[k,j] and not M[k+1, j + v]:
        M[k+1, j + v] = true
        V[i,k+1,j + v] = v
        P[i,k+1,j + v] = (I[j,k],k)
        I[j+v,k+1] = i
  For each j<S:
    If M[N-1, j]:
      A[j] = N-1
  M[0,v] = true
  I[v,0] = i
  ++i
Find first nonzero element in A[N-1,S..S+U] (or the first element with smallest\
  value or any other element that suits both minimization criteria)
Reconstruct elements of the subset using matrices V and P.

Time complexity is O(L*N*S), where L is the length of the list, N and S are given limits.

Space complexity is O(L*N*S).