traditional game playing 2 player adversarial (win => lose) based on search but... huge game...

8
traditional game playing • 2 player • adversarial (win => lose) • based on search • but... huge game trees can't be fully explored

Upload: lee-kelley

Post on 31-Dec-2015

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

traditional game playing

• 2 player

• adversarial (win => lose)

• based on search

• but... huge game trees can't be fully explored

Page 2: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

traditional game playing• 2 player• adversarial (win => lose)• fixed rules – no general world kn• based on search but...

huge game trees – can't be fully explored

why study them in AI?• core part of tools & techniques• adversary modelling is important

economics, contingency planning & other areas

Page 3: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

trad. game playing

• basics• minimax search routine

• depth 1st to fixed depth• different approaches to copy, cache, states

• static evaluation fn• assesses merit of game states for players• simple +/- numeric value

• alpha-beta pruning• std approach for reducing game trees

Page 4: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

alpha – beta … continued

• best & worst cases

• improving alpha-betasimple ordering fns

eg: captures => threats => moves

Page 5: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

other strategies (growth, etc)

• eval all nodes & extend treeheuristic growth

quiescence

plausibility effort

• use different eval fns at different stagesstrategy, performance, etc

• library moves (open game / end game)state representations

database lookup

Page 6: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

other strategies (pruning)

• eval all nodes & prune treeheuristic pruning

limiting breadth

futility cut-off

• caching stateswhen/why to cache

cache persistence=> library moves?

Page 7: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

minimum needs

1. a state representation

2. a static evaluation fn

3. a legal move generator

4. minimax

5. alpha-beta pruning?

Page 8: Traditional game playing 2 player adversarial (win => lose) based on search but... huge game trees can't be fully explored

uncertainty

1. chance – dice games, etc

2. incomplete kn – cards, etc