Database Reference
In-Depth Information
Definition 8.2 (Stochastic dominance). For two paths P 1 and P 2 , P 1 stochastically
dominates P 2 ,if w P 1 , the weight or P 1 , is smaller than w P 2 , the weight of P 2 ,in
stochastic order.
P opt stochastically dominates all paths in
P 2 ,if w P opt is smaller than any w P i
(1
m ) in stochastic order. An example is shown in Figure 8.5(b). P opt stochas-
tically dominates
i
. The following theorem shows that P opt can be used to
define a valid heuristic evaluation function.
B
,
D
,
E
Theorem 8.6 (Sufficient condition). Let
P
= {
P 2 1 ,...,
P 2 m }
be all paths between
2
v i and v in graph G. For a path P opt
=
v i
,...,
v
,ifP opt stochastically dominates all
paths in
P 2 , then
Δ (
v i ,
l
)=
f P 1 , e (
x 1 ,
y
) ×
f P opt (
x 2 )
x 1 +
y
+
x 2
l
is a valid heuristic evaluation function for P*.
Proof. Compare two paths P
P opt and P =
=
P 1 +
P 1 +
P 2 i
( P 2 i ∈P 2 ).
]= x l Pr [ w P 1 = x ] × Pr [ w P opt l x ]
Pr
[
w P
l
w P
]= x l Pr [ w P 1 = x ] × Pr [ w P 2 i l x ]
Pr
[
l
w P
Since Pr
[
w P opt
l
x
]
Pr
[
w P 2 i
l
x
]
,wehave Pr
[
w P
l
]
Pr
[
l
]
.
More than one P opt can be constructed. Here we present a simple three-step con-
struction method that ensures the resulting path is a stochastically dominating path
for
P 2 .
Step 1: Constructing the path
We find the path P 2 i
2 with the least number of edges. Let n be the num-
ber of edges in P 2 i . We construct n
in
P
1 virtual vertices v 1
,...,
v n 1 . Let P opt
=
v i ,
v 1 ,...,
v n 1 ,
v
. Thus, P opt has the least number of edges among all edges in
P 2 .
For example, in the probabilistic graph in Figure 2.6, there are two paths between
B and E : P 1 =
B
,
E
and P 2 =
B
,
D
,
E
. Since P 1 only contains one edge, the path
P opt should also contain only one edge.
Step 2: Assigning edge weights
Let
P 2 . We want to construct the weight of an edge e opt in
P opt such that e opt stochastically dominates all edges in
E
be the set of edges in
E
.
 
Search WWH ::




Custom Search