8

How does in the (famous Zilberstein) PR(uning) algorithm below the LP-dominate function get started: the first time it's called, D=∅ and the linear program deteriorates (i.e. no constraint equations)? 

procedure POINTWISE-DOMINATE(w, U)
...
3. return false
procedure LP-DOMINATE(w, U)
4. solve the following linear program variables: d, b(s) ∀s ∈ S
      maximize d
      subject to the constraints
        b · (w − u) ≥ d, ∀u ∈ U
        sum(b) = 1
5. if d ≥ 0 then return b
6. else return nil
procedure BEST(b, U )
...
12. return w
procedure PR(W)
13. D ← ∅
14. while W = ∅
15.   w ← any element in W
16.   if POINTWISE-DOMINATE(w, D) = true
17.      W ← W − {w}
18.   else
19.      b ← LP-DOMINATE(w, D)
20.      if b = nil then
21.         W ← W − {w}
22.      else
23.         w ← BEST(b, W)
24.         D ← D ∪ {w}
25.         W ← W − {w}
26. return D
Blaszard
  • 1,037
  • 3
  • 11
  • 25
stustd
  • 81
  • 1

1 Answers1

1

I think I found the solution. When in PR(W), D=∅, the weight is:

b[i] = 0 for { i | w[i]<max(w) },

and

b[i] = 1.0/max(w) for { i | w[i]==max(w) }.

Maxim
  • 1,957
  • 2
  • 16
  • 27
stustd
  • 111
  • 3