Visual clustering with artificial ants colonies

3 downloads 0 Views 146KB Size Report
64, avenue Jean Portalis, 37200 Tours, FRANCE. {labroche,monmarche,venturini}@ univ-tours.fr. In this paper, we propose a new approach inspired from the ...
!" -

# "

"$%&'' " ./

"()*+, 0

! " &

#

%$% !

#

%$'

!

)

, ' &

(

% +

*

* /.,& 0123

+ 0#2

%)

.

,

04 52 06 72 .

! .

!

082

. ,

% )

! . ! & *

+ '

&

9

# / 3

. ,

4 5

. , ) *

6

)

.

/

3

" /

3 .

' ; 0< =2

:

'

. ,

! *



#

%$[> 1] •

" -

. !

1



/ '

/ '

# ! 6>3 $ 2 / ⋅3

3

'/2 / ⋅33

9 ←



.

#

%$2

2

/ ⋅3 +

' /2

/ ⋅33

#

.

#$% (

(

> 2 #

%$3 '

Algorithm: Meeting(Ant i, Ant j) 1. D ← Compute the Euclidean distance between Labeli and Labelj 2. If (D

%

/

!

3

/

#3

? -

*

? )

39) ← 2

/1

1

3

3−

%

(

. ,

Algorithm: Visual AntClust (data set with N objects, Niter iterations) 1. 2. 3. 4. 5. 6. 7. 8.

Initialise the N ants according to parameters described in section 2.2 While (Niter iterations are not reached) do Draw N ants in the odour space For c ( B

7

. ,

. ,

@%

. ,

%

1

. ,

# ,

1 C

.

1>>

>

4 . .

5 , 6>>

.

*

1 *

. ,

#"

1>>>

#

%$.

1

$ . , ! ' . ,

@% %

(

. , . ,

( +

! 1 #

"C &' F

$ B

D +

+A .

E ) 1=== G$

A

9* H"

.

A

B

A

1==7 4 5

E , J146 E I

6 3 7

" B

8

< =

#>>#3 C LM , L

( : ;" ! & 514%5#8 #>>1 A C E :. J45 516%5#5 #>>1 A :& ; . /$ * K 7#7%746 1=== C * :$ ; , A . 9* . , 1==5 6>1%6>< :. ; B 16 " , . * 456%45= #>># "( * A C 1==> C LM : ) /, ;C " A 1= 1#4%145 1=