4.4. GENERAL WIN PREDICTION IN GVGAI 67
4.4 GENERAL WIN PREDICTION IN GVGAI
4.4.1 GAME PLAYING AGENTS AND FEATURES
Random Search (RS)
L
H
C
H
f D score C
8
<
:
H
C
;
H
;
L 10 30
90
Rolling Horizon Evolutionary Algorithm (RHEA)
P D 2 L D 8 P D
10 L D 14 L P
68 4. FRONTIERS OF GVGAI PLANNING
Monte Carlo Tree Search (MCTS)
3 W D 2 L D 8 W D 10 L D 10
W D 10 L D 14 W L
14 100
5 20
900
100
100
L P
L
# Algorithm
Victory Rate
(Standard Error)
1 10-14-EA-Shift 26.02% (2.11)
2 2-8-EA-Shift 24.54% (2.00)
3 10-RS 24.33% (2.13)
4 14-MCTS 24.29% (1.74)
5 10-MCTS 24.01% (1.65)
6 10-14-EA-MCTS 23.99% (1.80)
7 2-8-EA-MCTS 23.98% (1.73)
8 2-8-EA-ALL 23.95% (1.98)
9 8-MCTS 23.42% (161)
10 10-14-RHEA 23.23% (2.08)
11 10-14-EA-ALL 22.66% (2.02)
12 30-RS 22.49% (2.02)
13 2-8-RHEA 18.33% (1.77)
14 90-RS 16.31% (1.67)
1
Current game score
4.4. GENERAL WIN PREDICTION IN GVGAI 69
2
Convergence
3
Positive rewards
4
Negative rewards
5
Success
6
Danger
7
Improvement
8
Decisiveness
9
Options exploration
10
Fitness distribution
11
Success distribution
12
Danger distribution
H.X/ D
N 1
X
iD0
p
i
log
2
p
i
:
2
8
9
10
11
12
t
5
6
11
12
https://github.com/rdgain/ExperimentData/tree/GeneralWinPred-CIG-18
2:5
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.