52 3. PLANNING IN GVGAI
! D 15 SD
D 0:05
SD
C
D 0:4
ı
3.5.2 RHEA ENHANCEMENTS
Population initialization
3.5. ROLLING HORIZON EVOLUTIONARY ALGORITHMS 53
Algorithm 3.4
Input: t
Input: Ld
Input: L
Requires: !
Requires: SD
f
Ld
L
Requires: SD
C
f
Ld
L
Requires: M
D
Requires: MIN
L
L
Requires: MAX
L
L
procedure Ld; t; L
if t mod ! D 0 then
if Ld D null then
f
Ld
SD
f
Ld
else
f
Ld
ı.Ld/
end if
if f
Ld
< SD
then
L L C M
D
else if f
Ld
> SD
C
then
L L M
D
end if
.L; MIN
L
; MAX
L
/
end if
procedure L; MIN
L
; MAX
L
if L < MIN
L
then
L MIN
L
else if L > MAX
L
then
L
MAX
L
end if
return L
54 3. PLANNING IN GVGAI
P L
Bandit-based mutation
C D
p
2
Q.s; a/
3.5. ROLLING HORIZON EVOLUTIONARY ALGORITHMS 55
Statistical tree
Individual Evaluation
Individual
Rollout
Backpropagation
Evaluate State
Population
↑
(crossover)
↑
mutate
↑
select individual
a
0
i
0
a
0,
a
1, …,
a
n
i
1
a
0,
a
1, …,
a
n
i
n
a
0,
a
1, …,
a
n
a
1
a
n
•
•
•
•
•
•
P L
Shiftbuffer
56 3. PLANNING IN GVGAI
MonteCarlorolloutevaluation
r
L
r
N
r
L
f D
P
r
N
nD1
V
.
s
n
/
r
N
:
L
r
L
L
r
N
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.