일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | ||||||
2 | 3 | 4 | 5 | 6 | 7 | 8 |
9 | 10 | 11 | 12 | 13 | 14 | 15 |
16 | 17 | 18 | 19 | 20 | 21 | 22 |
23 | 24 | 25 | 26 | 27 | 28 | 29 |
30 | 31 |
- OPENHACK
- Qwik Start
- 머신러닝
- rl
- Speech
- 코딩테스트
- 스터디 잼
- kakao
- Conference
- re:Invent
- CSIP
- SageMaker
- 구글스타트업캠퍼스
- 뉴비톤
- aws
- Ground Truth
- Tensorflow 2.0
- entity
- Backend.AI
- ainize
- Community Day
- SW중심대학
- GDG Campus
- API
- 해커톤
- gcp
- BOAZ
- Open Hack
- seq2seq
- 오픈소스해커톤
- Today
- Total
pizzaplanet
Search Problems 본문
본 강의는 UC Berkeley의 AI 강의를 참고하여 진행됩니다.
Pacman을 예시로 사용하여 Search Problems의 3가지 구성요소를 살펴보자. (pac coin CPR좀..)
A State space
State space는 각 상황을 나타내는 state의 집합이다.
state는 x,y에 위치한 food의 on, ff의 상태이다.
위 이미지에서 간략하게 state space를 계산한다면 3 X 3 크기에 food가 on, off로 나뉘므로 으로 볼 수 있다.
물론 더 들어가서 팩맨의 위치, 팩맨의 방향까지 고려할 수 있지만 용어 및 개념파악을 위한 것이므로 더 자세히 들어가지는 말자.
A successor function(with actions, costs)
왼쪽 현재 state에서 "N",1.0이 실행되면 우상단의 이미지와 같이 팩맨이 North 방향으로 cost 1 만큼 이동한다.
- A start state and a goal test
시작점에서 종료지점까지로 변환되는 행동(계획)의 시퀀스를 솔루션이라 부른다.
이제 3가지 구성요소를 살펴보았으니 실습을 통해 감을 익혀보자.
States, Actions, Successor, Goal test 구하기
Problem |
길찾기 |
모든 점 먹기 |
States |
(x,y) location |
{(x,y), dot booleans} |
Actions |
NSEW |
NSEW |
Successor |
update location only |
Update location and possibly a dot boolean |
Goal test |
(x,y)=END |
dots all false |
같은 상황에서 골지점에 따라 state, successor가 달라질 수 있는 것이 포인트!
State Space Size 구하기
Map size: 5 * 6
Agent facing: 4 [NSEW]
Food Count: 30 [Map size]
Agent Positions: 120 [Map size * Agent facing]
Ghost positions: 12
World states: [Agent Positions * 2^Food count * Ghost positions^2 * Agent facing]
States for pathing: 120
States for eat-all-dots: 120*2^30
'Univ. > Artificial intelligence lecture' 카테고리의 다른 글
Tree Search (0) | 2018.04.24 |
---|---|
State Space Graphs vs. Search Trees (0) | 2018.04.15 |
Agents (0) | 2018.04.04 |
Other places you will meet AI (0) | 2018.04.04 |
What makes an AI problem easy or hard? (0) | 2018.04.04 |