+ All Categories
Home > Documents > Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper...

Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper...

Date post: 21-Dec-2015
Category:
View: 216 times
Download: 0 times
Share this document with a friend
Popular Tags:
23
Go Meeting Talk 1 Generation of Patterns with External Conditions for the Game of Go Paper presentation http://www.ai.univ-paris8.fr/ ~cazenave/acg9.pdf Author: Tristan Cazenave Presentation by: Adi Botea
Transcript
Page 1: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 1

Generation of Patterns with External Conditions for the

Game of Go

Paper presentation

http://www.ai.univ-paris8.fr/~cazenave/acg9.pdf

Author: Tristan Cazenave

Presentation by: Adi Botea

Page 2: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 2

Outline

Motivation Go Patterns with External Conditions

– What they are– Representation– Generation

Experimental Results Conclusion

Page 3: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 3

Motivation

Heuristic Search needs enhancements (e.g., transposition tables)

Go Patterns can be a powerful enhancement to Heuristic Search

Humans can learn from such databases

Page 4: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 4

Go Patterns with External Conditions

Images reproduced from the paper

Page 5: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 5

Go Patterns with External Conditions

Images reproduced from the paper

Page 6: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 6

Go Patterns with External Conditions

Images reproduced from the paper

Page 7: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 7

Go Patterns with External Conditions External Conditions:

– It is the number of external liberties of groups that matters

– Used in many Go programs BUT…– Now they are generated automatically

Page 8: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 8

Go Patterns with External Conditions

Image reproduced from the paper

Page 9: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 9

Representation

MaxNrLibertyIfEnemy: no conditions, 0, 1 MinNrLibertyIfFriend:no conditions, 1, 2 MaxNrLiberty:no conditions, 0, 1 MinNrLiberty:no conditions, 1, 2

9x9x3x3 = 729 rules

Image reproduced from the paper

Page 10: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 10

RepresentationMoves for White:

Images reproduced from the paper

Page 11: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 11

Included Patterns

Images reproduced from the paper

Page 12: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 12

Included Patterns

Images reproduced from the paper

Page 13: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 13

Number of Possible Patterns

Image reproduced from the paper

Page 14: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 14

Generation

Each pattern coded as a 32-bit integer Save memory:

– Only a few rules lead to win or winning positions

– Different sets of conditions can be associated with a pattern

Page 15: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 15

Generation

Page 16: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 16

Generation

Page 17: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 17

Generation

Further optimizations:– Unmove only the last deduced patterns– Order of Test and Cut– Rule-Coverage Reductions

Image reproduced from the paper

Page 18: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 18

Results – Eyes on the side

Images reproduced from the paper

Page 19: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 19

Results – Life in the Corner

Images reproduced from the paper

Page 20: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 20

Results

Image reproduced from the paper

Page 21: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 21

Results

Image reproduced from the paper

Page 22: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 22

Results

Image reproduced from the paper

Page 23: Go Meeting Talk1 Generation of Patterns with External Conditions for the Game of Go Paper presentation cazenave/acg9.pdf.

Go Meeting Talk 23

Conclusion

Go Patterns with External Conditions:– What are they?– How to represent them?– How to generate them?– Results


Recommended