T
E
L
KO
M
N
I
KA
T
e
lec
om
m
u
n
icat
ion
,
Com
p
u
t
i
n
g,
E
lec
t
r
on
ics
an
d
Cont
r
ol
Vol.
18
,
No.
1
,
F
e
br
ua
r
y
2020
,
pp.
562
~
570
I
S
S
N:
1693
-
6930,
a
c
c
r
e
dit
e
d
F
ir
s
t
G
r
a
de
by
Ke
me
nr
is
tekdikti
,
De
c
r
e
e
No:
21/E
/KP
T
/2018
DO
I
:
10.
12928/
T
E
L
KO
M
NI
KA
.
v18i1.
13497
562
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
i
-
E
c
la
t
:
p
e
r
f
or
m
a
n
c
e
e
n
h
a
n
c
e
m
e
n
t
of
E
c
la
t
vi
a i
n
c
r
e
m
e
n
t
al
ap
p
r
oac
h
i
n
f
r
e
q
u
e
n
t
i
t
e
m
s
e
t
m
in
in
g
Wan
Ae
z
wani
Wan
Abu
B
ak
ar
1
,
M
u
s
t
af
a
M
an
2
,
M
ah
ad
i
M
an
3
,
Z
ail
an
i
Abd
u
ll
ah
4
1
Facu
l
t
y
o
f
I
n
fo
rma
t
i
c
s
an
d
C
o
mp
u
t
i
n
g
,
U
n
i
v
er
s
i
t
i
Su
l
t
a
n
Z
ai
n
al
A
b
i
d
i
n
,
Mal
a
y
s
i
a
2
,3
Sch
o
o
l
o
f
In
f
o
rmat
i
cs
an
d
A
p
p
l
i
e
d
Mat
h
ema
t
i
c
s
,
U
n
i
v
ers
i
t
i
Mal
a
y
s
i
a
T
ere
n
g
g
an
u
,
Mal
ay
s
i
a
4
Facu
l
t
y
o
f
E
n
t
rep
re
n
eu
r
s
h
i
p
an
d
Bu
s
i
n
es
s
(FE
B)
,
Cen
t
r
e
o
f
Co
mp
u
t
i
n
g
an
d
I
n
fo
rma
t
i
c
s
(CCI),
U
n
i
v
er
s
i
t
i
Mal
a
y
s
i
a
K
e
l
an
t
an
,
Ci
t
y
Camp
u
s
,
Mal
a
y
s
i
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
J
ul
4
,
2019
R
e
vis
e
d
J
ul
2
8
,
2020
Ac
c
e
pted
Oc
t
22,
2019
O
n
e
ex
amp
l
e
o
f
t
h
e
s
t
a
t
e
-
of
-
the
-
ar
t
v
ert
i
cal
ru
l
e
mi
n
i
n
g
t
ech
n
i
q
u
e
i
s
cal
l
ed
eq
u
i
v
a
l
en
ce
c
l
as
s
t
ran
s
fo
rma
t
i
o
n
(
E
cl
a
t
)
a
l
g
o
ri
t
h
m.
N
ei
t
h
er
h
o
ri
z
o
n
t
al
n
o
r
v
ert
i
cal
d
a
t
a
fo
rma
t
,
b
o
t
h
are
s
t
i
l
l
s
u
fferi
n
g
fro
m
t
h
e
h
u
g
e
memo
r
y
co
n
s
u
m
p
t
i
o
n
.
In
res
p
o
n
s
e
t
o
t
h
e
p
ro
m
i
s
i
n
g
res
u
l
t
s
o
f
mi
n
i
n
g
i
n
a
h
i
g
h
er
v
o
l
u
me
o
f
d
at
a
fr
o
m
a
v
er
t
i
ca
l
fo
rma
t
,
an
d
t
a
k
i
n
g
c
o
n
s
i
d
era
t
i
o
n
o
f
d
y
n
ami
c
t
ran
s
act
i
o
n
o
f
d
at
a
i
n
a
d
at
a
b
as
e,
t
h
e
res
earch
p
ro
p
o
s
e
s
a
p
erfo
rman
ce
en
h
a
n
cemen
t
o
f
E
cl
a
t
al
g
o
r
i
t
h
m
t
h
at
re
l
i
e
s
o
n
i
n
creme
n
t
a
l
ap
p
ro
ac
h
cal
l
ed
an
In
creme
n
t
a
l
-
E
c
l
at
(
i
-
E
c
l
at
)
a
l
g
o
ri
t
h
m.
M
o
t
i
v
a
t
ed
fr
o
m
t
h
e
fa
s
t
i
n
t
ers
ec
t
i
o
n
i
n
E
cl
a
t
,
t
h
i
s
al
g
o
r
i
t
h
m
o
f
p
erf
o
rman
ce
en
h
an
ceme
n
t
a
d
o
p
t
s
v
i
a
my
s
t
ru
c
t
u
re
d
q
u
er
y
l
an
g
u
a
g
e
(My
S
Q
L
)
d
at
a
b
as
e
ma
n
ag
eme
n
t
s
y
s
t
em
(D
BMS)
as
i
t
s
p
l
a
t
fo
rm.
It
s
erv
es
as
t
h
e
as
s
o
c
i
at
i
o
n
ru
l
e
mi
n
i
n
g
d
at
a
b
as
e
en
g
i
n
e
i
n
t
es
t
i
n
g
b
en
c
h
mark
freq
u
en
t
i
t
ems
e
t
mi
n
i
n
g
(FIMI)
d
at
a
s
et
s
fro
m
o
n
l
i
n
e
rep
o
s
i
t
o
r
y
.
T
h
e
My
S
Q
L
D
BMS
i
s
ch
o
s
en
i
n
o
rd
er
t
o
red
u
ce
t
h
e
p
r
ep
ro
ce
s
s
i
n
g
s
t
a
g
es
o
f
d
at
a
s
et
s
.
T
h
e
ex
p
eri
me
n
t
a
l
res
u
l
t
s
i
n
d
i
cat
e
t
h
a
t
t
h
e
p
ro
p
o
s
e
d
al
g
o
r
i
t
h
m
o
u
t
p
erf
o
rms
t
h
e
t
ra
d
i
t
i
o
n
a
l
E
cl
a
t
w
i
t
h
1
7
%
b
o
t
h
i
n
ch
e
s
s
an
d
T
1
0
I4
D
1
0
0
K
,
6
9
%
i
n
mu
s
h
ro
o
m,
5
%
an
d
8
%
i
n
p
u
m
s
b
_
s
t
ar
an
d
ret
a
i
l
d
at
a
s
et
s
.
T
h
u
s
,
amo
n
g
fi
v
e
(5
)
d
en
s
e
an
d
s
p
ar
s
e
d
a
t
as
e
t
s
,
t
h
e
av
era
g
e
p
erfo
rman
ce
o
f
i
-
E
cl
a
t
i
s
co
n
c
l
u
d
ed
t
o
b
e
2
3
%
b
et
t
er
t
h
a
n
E
c
l
at
.
K
e
y
w
o
r
d
s
:
As
s
oc
iation
r
ule
mi
ning
(
AR
M
)
De
ns
e
da
tas
e
t
E
c
lat
I
nc
r
e
menta
l
-
e
c
lat
(
i
-
E
c
lat)
S
pa
r
s
e
da
tas
e
t
Ve
r
ti
c
a
l
f
or
mat
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e
.
C
or
r
e
s
pon
din
g
A
u
th
or
:
W
a
n
Ae
z
wa
ni
W
a
n
Abu
B
a
ka
r
,
F
a
c
ult
y
of
I
nf
or
matics
a
nd
C
omput
ing,
Unive
r
s
it
i
S
ult
a
n
Z
a
inal
Abidin
,
B
e
s
ut
C
a
mpus
,
22200
B
e
s
ut
T
e
r
e
ngga
nu,
609
-
6993054,
M
a
lays
ia.
E
mail:
wa
na
e
z
wa
ni@uni
s
z
a
.
e
du.
my
1.
I
NT
RODU
C
T
I
ON
Da
ta
mi
ning
(
DM
)
is
the
r
e
s
e
a
r
c
h
a
r
e
a
whe
r
e
the
huge
da
tas
e
t
in
da
taba
s
e
a
nd
da
ta
r
e
pos
it
or
y
a
r
e
be
ing
s
c
our
e
d
a
nd
mi
ne
d
to
f
ind
nove
l
a
nd
us
e
f
ul
pa
tt
e
r
n.
As
s
oc
iation
a
na
lys
is
is
one
of
the
f
our
(
4)
c
or
e
da
ta
mi
ning
tas
ks
be
s
ides
c
lus
ter
a
na
lys
is
,
pr
e
dictive
m
ode
ll
ing
a
nd
a
nomaly
de
tec
ti
on.
T
he
tas
k
of
a
s
s
oc
iation
r
ule
mi
ning
is
to
dis
c
ove
r
i
f
ther
e
e
xis
t
the
f
r
e
que
nt
it
e
ms
e
t
o
r
pa
tt
e
r
n
in
da
taba
s
e
a
nd
i
f
a
ny,
a
n
in
ter
e
s
ti
ng
r
e
lations
hip
be
tw
e
e
n
thes
e
f
r
e
que
nt
it
e
ms
e
ts
c
a
n
r
e
ve
a
l
a
ne
w
pa
tt
e
r
n
a
na
lys
is
f
o
r
the
f
utur
e
de
c
is
ion
making.
I
t
is
a
f
unda
menta
l
pa
r
t
of
many
da
ta
mi
ning
a
ppli
c
a
ti
ons
including
mar
ke
t
ba
s
ke
t
a
na
lys
is
,
we
b
li
nk
a
na
lys
is
,
ge
nome
a
na
lys
is
a
nd
mol
e
c
ular
f
r
a
gment
mi
nin
g.
I
n
m
ini
ng
f
r
e
que
nt
it
e
ms
e
ts
,
two
(
2)
main
s
e
a
r
c
hing
st
r
a
tegie
s
c
ould
be
a
ppli
e
d
i
.
e
.
Hor
izonta
l
f
or
mat
(
br
e
a
dth
f
i
r
s
t
s
e
a
r
c
h)
or
ve
r
ti
c
a
l
f
or
mat
(
de
pth
f
ir
s
t
s
e
a
r
c
h)
.
T
he
s
e
a
r
c
hing
s
tr
a
tegy
in
mi
ning
mi
ght
give
s
ign
if
ica
nt
e
f
f
e
c
t
to
ove
r
a
ll
m
ini
ng
p
r
oc
e
s
s
.
F
in
ding
f
r
e
que
nt
it
e
ms
e
ts
or
pa
tt
e
r
ns
is
a
big
c
ha
ll
e
nge
a
nd
ha
s
a
s
tr
o
ng
a
nd
long
-
s
tanding
tr
a
dit
ion
in
da
ta
mi
ning
[
1]
s
i
nc
e
da
ta
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
i
-
E
c
lat
:
pe
r
for
manc
e
e
nhanc
e
me
nt
of
E
c
lat
v
ia
incr
e
me
ntal
appr
oac
h…
(
W
an
A
e
z
w
ani
W
an
A
bu
B
ak
ar
)
563
is
incr
e
a
s
ing
r
a
pidl
y
in
volum
e
[
2]
.
T
he
di
f
f
iculti
e
s
a
r
is
e
whe
n
ne
w
da
ta
is
a
dde
d,
the
na
ïve
a
ppr
oa
c
h
is
to
r
e
r
un
the
whole
mi
ning
a
lgor
it
hm
[
3]
on
the
whole
da
tas
e
ts
.
T
hus
,
th
is
pr
oc
e
s
s
r
e
s
ult
s
in
huge
main
a
nd
c
a
c
he
memor
y
c
ons
umpt
ion
[
4]
.
T
o
the
be
s
t
of
our
knowle
dge
,
ther
e
a
r
e
thr
e
e
(
3)
ba
s
ic
f
r
e
que
nt
i
tems
e
t
mi
ning
a
nd
be
s
t
-
known
a
lgor
it
hms
,
Apr
ior
i
[
1
,
2
]
that
li
e
s
in
h
or
iz
ontal
f
o
r
mat,
F
P
-
Gr
owth
[
3
]
a
nd
E
c
lat
[
4]
th
a
t
li
e
s
in
ve
r
ti
c
a
l
f
or
mat
.
I
n
thi
s
pa
pe
r
,
we
f
oc
us
on
v
e
r
ti
c
a
l
f
or
mat
by
look
ing
de
e
pe
r
on
e
quivale
n
c
e
c
las
s
tr
a
ns
f
o
r
mation
(
E
c
lat)
a
lgo
r
it
hm
[
4
]
a
nd
E
c
lat
-
va
r
iants
a
lgor
it
hm
in
[5
-
7]
a
s
the
r
e
s
e
a
r
c
h
ba
s
e
models
.
W
e
pr
opos
e
a
n
incr
e
menta
l
a
ppr
oa
c
h
that
ba
s
e
d
on
ve
r
ti
c
a
l
E
c
lat
a
lgor
it
hm
a
nd
na
med
it
a
s
incr
e
menta
l
E
c
lat
o
r
i
-
E
c
lat
to
im
pr
ove
the
pe
r
f
or
manc
e
of
memor
y
.
T
h
e
r
e
s
pons
e
ti
me
is
mea
s
ur
e
d
in
ti
me
e
xe
c
uti
on
pe
r
s
e
c
onds
.
2.
B
ASI
C
P
RI
NC
I
P
L
E
S
2.
1.
As
s
oc
iat
ion
r
u
le
I
n
the
r
e
s
t
of
thi
s
s
e
c
ti
on,
let
be
the
it
e
m
ba
s
e
whe
r
e
=
{
1
,
2
,
.
.
.
,
}
,
for
>
0
r
e
f
e
r
s
to
the
s
e
t
of
li
ter
a
ls
c
a
ll
e
d
a
s
e
t
of
m
it
e
ms
.
I
tems
may
be
a
pr
oduc
t
,
s
e
r
vice
,
a
c
ti
on
o
r
a
tom
.
A
s
e
t
=
{
1
,
.
.
.
,
}
⊆
is
c
a
l
led
a
n
it
e
ms
e
t
or
a
k
-
it
e
ms
e
t
if
it
c
on
tains
it
e
ms
.
A
tr
a
ns
a
c
ti
on
ove
r
is
a
c
ouple
o
f
=
(
,
)
w
he
r
e
is
c
a
ll
e
d
a
tr
a
ns
a
c
ti
on
id
e
nti
f
ier
a
nd
is
a
n
it
e
ms
e
t.
A
tr
a
ns
a
c
ti
on
=
(
,
)
is
s
a
id
to
s
uppor
t
a
n
it
e
ms
e
t
⊆
if
⊆
.
A
tr
a
ns
a
c
ti
on
da
taba
s
e
is
a
s
e
t
of
tr
a
ns
a
c
ti
on
ove
r
.
A
ti
ds
e
t
of
a
n
it
e
ms
e
t
in
is
de
f
ined
a
s
a
s
e
t
of
tr
a
ns
a
c
ti
on
identif
ier
s
in
that
s
uppor
t
whe
r
e
(
)
=
{
|
(
,
)
∈
,
⊆
}
.
S
uppor
t
o
f
a
n
it
e
ms
e
t
in
is
the
c
a
r
dinali
ty
o
f
it
s
ti
ds
e
t
s
uc
h
that
s
uppor
t
of
is
the
number
o
f
t
r
a
ns
a
c
ti
ons
c
ontaining
in
,
whe
r
e
(
)
=
|
(
)
|
.
I
ll
us
tr
a
ti
on
of
s
uppor
t
-
c
onf
idenc
e
f
r
a
mew
or
k
is
given
a
s
be
low:
-
T
he
s
uppor
t
o
f
r
u
le
⇒
is
the
f
r
a
c
ti
on
of
tr
a
ns
a
c
ti
ons
i
n
da
taba
s
e
,
D
c
ontaining
both
X
a
nd
Y.
(
⇒
)
=
∪
|
|
-
T
he
c
onf
idenc
e
of
r
ule
⇒
is
the
f
r
a
c
ti
on
of
tr
a
ns
a
c
ti
ons
in
D
c
ontaining
X
that
a
ls
o
c
ontain
Y.
(
⇒
)
=
(
∪
)
(
)
A
r
ule
is
fr
e
que
nt
i
f
it
s
s
uppor
t
is
gr
e
a
ter
than
mi
ni
mum
s
uppor
t
(
mi
n_s
upp)
th
r
e
s
hold.
T
he
r
ules
whic
h
s
a
ti
s
f
y
mi
nim
um
c
on
f
idenc
e
(
mi
n_c
onf
)
thr
e
s
hold
is
c
a
ll
e
d
s
tr
ong
r
ule
whe
r
e
_
a
nd
_
a
r
e
us
e
r
s
pe
c
if
ied
va
lues
.
An
a
s
s
oc
iation
r
ule
is
c
ons
ider
e
d
int
e
r
e
s
ti
ng
if
it
s
a
ti
s
f
ies
both
mi
n_s
upp
a
nd
mi
n_c
onf
thr
e
s
holds
[
8]
.
2.
1.
De
f
in
i
t
ion
s
De
f
ini
ti
on
1
.
Give
n
a
tr
a
ns
a
c
ti
on
da
taba
s
e
T
ove
r
a
n
it
e
m
ba
s
e
B
a
nd
a
mi
nim
a
l
s
uppor
t
thr
e
s
hold,
s
m
i
n
.
T
he
s
e
t
of
a
ll
f
r
e
que
nt
it
e
ms
e
ts
is
de
noted
by:
F
(
T
,
s
m
i
n
)
=
{
X
⊆
B
|
s
u
p
(
X
)
≥
s
m
i
n
}
De
f
ini
ti
on
2
.
(
C
a
ndidate
i
tems
e
t)
:
Give
n
a
t
r
a
ns
a
c
ti
on
da
taba
s
e
T
with
a
m
ini
mum
s
uppor
t
th
r
e
s
hold,
S
m
i
n
a
nd
a
lgo
r
it
hm
f
or
f
r
e
que
nt
it
e
ms
e
t
mi
ning
of
F
(
T
,
s
m
i
n
)
,
a
n
it
e
ms
e
t
X
is
c
a
ll
e
d
c
a
ndidate
it
e
ms
e
t
if
the
a
lgor
it
hm
e
va
luate
s
if
X
is
f
r
e
que
nt
or
not.
De
f
ini
ti
on
3.
(
I
nter
s
e
c
ti
on)
:
L
e
t
A
a
nd
B
be
s
e
ts
.
T
he
int
e
r
s
e
c
ti
on
of
s
e
t
A
a
nd
B
is
de
noted
by
A
∩
B
is
the
s
e
t
c
ontaining
thos
e
e
leme
nts
in
both
A
a
nd
B
s
uc
h
that
∩
=
{
|
∈
ʌ
∈
}
De
f
ini
ti
on
4.
(
Di
f
f
e
r
e
nc
e
s
e
t)
:
L
e
t
A
a
nd
B
be
s
e
ts
.
T
he
dif
f
e
r
e
nc
e
of
A
a
nd
B
is
de
noted
b
y
A
–
B
,
is
the
s
e
t
c
ontaining
thos
e
e
leme
nts
that
a
r
e
in
A
but
not
in
B
.
T
he
di
f
f
e
r
e
nc
e
of
A
a
nd
B
is
a
ls
o
c
a
ll
e
d
the
c
ompl
e
ment
of
B
with
r
e
s
pe
c
t
to
A
s
uc
h
that
A
−
B
=
{
x
|
x
∈
A
ʌ
x
∉
B
}
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
562
-
570
564
3.
T
RA
DI
T
I
ONAL
E
C
L
AT
E
c
lat
is
the
a
bbr
e
viation
of
E
quivale
nc
e
C
las
s
T
r
a
ns
f
or
mation
.
I
ts
main
ope
r
a
ti
on
is
int
e
r
s
e
c
ti
on
of
ti
ds
e
ts
,
thus
the
s
ize
o
f
ti
ds
e
ts
is
one
of
the
main
f
a
c
tor
s
a
f
f
e
c
ti
ng
the
r
unning
t
im
e
a
nd
memor
y
us
a
ge
of
E
c
lat
[
9
,
10
]
.
T
he
bigger
ti
ds
e
ts
a
r
e
,
the
mor
e
ti
me
a
nd
memor
y
a
r
e
ne
e
de
d
[
11]
.
E
c
lat
us
e
s
pr
e
f
ix
-
ba
s
e
d
e
quivale
nc
e
r
e
lation,
θ1
a
long
with
b
ott
om
up
s
e
a
r
c
h.
I
t
e
numer
a
tes
a
ll
f
r
e
que
nt
it
e
ms
e
t
s
.
T
he
r
e
a
r
e
two
main
s
teps
:
c
a
ndidate
ge
ne
r
a
ti
on
a
nd
pr
u
ning.
I
n
c
a
ndidate
ge
ne
r
a
ti
on,
e
a
c
h
k
-
it
e
ms
e
t
c
a
ndidate
is
ge
ne
r
a
ted
f
r
om
two
f
r
e
que
nt
(
k
-
1)
-
it
e
ms
e
ts
a
nd
it
s
s
uppor
t
is
c
ounted
,
if
it
s
s
uppor
t
is
lowe
r
than
the
t
hr
e
s
hold,
then
it
will
be
d
is
c
a
r
de
d,
other
wis
e
it
is
f
r
e
que
nt
it
e
ms
e
ts
a
nd
us
e
d
to
ge
ne
r
a
te
(
k+
1)
-
it
e
ms
e
ts
.
E
c
l
a
t
s
tar
ts
with
pr
e
f
ix
{}
a
nd
the
s
e
a
r
c
h
tr
e
e
is
a
c
tually
the
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
.
T
o
divi
de
the
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
,
it
picks
t
he
pr
e
f
ix
{a
},
ge
ne
r
a
te
the
c
or
r
e
s
ponding
e
quivale
nc
e
c
las
s
a
nd
doe
s
f
r
e
que
nt
it
e
ms
e
t
mi
ning
in
the
s
ub
tr
e
e
of
a
ll
it
e
ms
e
ts
c
ont
a
ini
ng
{a
},
in
thi
s
s
ub
tr
e
e
it
divi
de
s
f
ur
ther
int
o
two
s
ub
tr
e
e
s
by
picking
the
pr
e
f
ix
{a
b}
:
the
f
ir
s
t
s
ub
tr
e
e
c
ons
is
ts
of
a
ll
it
e
ms
e
t
c
ontaining
{a
b},
the
other
c
ons
is
ts
of
a
ll
it
e
ms
e
ts
c
ontaining
{a
}
but
not
{b},
a
nd
thi
s
pr
oc
e
s
s
is
r
e
c
ur
s
ive
unti
l
a
l
l
it
e
ms
e
ts
in
th
e
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
a
r
e
vis
it
e
d
[
12]
.
3.
1.
Cand
i
d
at
e
ge
n
e
r
at
ion
an
d
p
r
u
n
i
n
g
I
n
c
a
ndidate
ge
ne
r
a
ti
on,
e
a
c
h
k
-
it
e
ms
e
t
c
a
ndidate
is
ge
ne
r
a
ted
f
r
om
two
f
r
e
que
nt
(k
-
1)
-
it
e
ms
e
ts
a
nd
it
s
s
uppor
t
is
c
ounted
[
13
,
14
]
.
I
f
it
s
s
uppor
t
is
lowe
r
th
a
n
the
th
r
e
s
hold,
then
it
will
be
pr
une
d
or
dis
c
a
r
de
d.
Othe
r
wis
e
,
it
is
f
r
e
que
nt
it
e
ms
e
ts
a
nd
us
e
d
to
ge
ne
r
a
te
(
k+
1)
-
it
e
ms
e
ts
.
S
ince
E
c
lat
us
e
s
ve
r
ti
c
a
l
layout,
c
ounti
ng
s
uppor
t
is
tr
ivi
a
l
.
De
pth
-
f
ir
s
t
s
e
a
r
c
hing
s
tr
a
tegy
is
done
whe
r
e
it
s
tar
ts
with
f
r
e
que
nt
it
e
ms
in
the
it
e
m
ba
s
e
a
nd
then
2
-
it
e
ms
e
ts
f
r
om
1
-
it
e
ms
e
ts
,
3
-
it
e
ms
e
ts
f
r
om
2
-
it
e
ms
e
ts
a
nd
s
o
on.
F
or
e
xa
mpl
e
,
{
}
=
{
}
∪
{
}
,
{
}
a
nd
{
}
a
r
e
pa
r
e
nt
o
f
{
}
.
T
o
a
void
ge
ne
r
a
ti
ng
dupli
c
a
te
it
e
ms
e
ts
,
(
k
-
1
)
-
it
e
ms
e
ts
a
r
e
s
or
ted
in
s
ome
or
de
r
s
.
T
o
ge
ne
r
a
te
a
ll
pos
s
ibl
e
k
-
it
e
ms
e
ts
f
r
om
a
s
e
t
of
(
−
1
)
-
it
e
ms
e
ts
s
ha
r
ing
(
−
2
)
it
e
ms
,
union
ope
r
a
ti
on
is
c
onduc
ted
o
f
a
(
−
1
)
-
it
e
ms
e
ts
with
the
it
e
ms
e
ts
that
s
tand
be
hind
it
in
the
s
or
ted
or
de
r
,
a
nd
thi
s
pr
oc
e
s
s
take
s
plac
e
f
or
a
ll
(
−
1
)
-
it
e
ms
e
ts
e
xc
e
pt
the
las
t
one
.
3.
2.
E
q
u
ival
e
n
c
e
c
las
s
An
e
quivale
nc
e
c
las
s
=
{
(
1
,
(
1
∪
)
)
,
…
,
(
,
(
∪
)
)
|
}
,
c
ons
ider
ing
the
s
e
t
{
1
,
…
,
}
a
s
a
n
it
e
m
ba
s
e
,
it
will
ha
ve
a
tr
e
e
of
it
e
ms
e
ts
ove
r
thi
s
it
e
m
ba
s
e
a
nd
if
the
pr
e
f
ix
P
is
a
ppe
nde
d
to
a
ll
it
e
ms
e
ts
in
thi
s
ne
w
tr
e
e
,
it
wi
ll
ha
ve
a
s
e
t
of
a
ll
it
e
ms
e
ts
s
ha
r
ing
the
pr
e
f
ix
in
the
s
e
a
r
c
h
tr
e
e
ove
r
the
i
tem
ba
s
e
.
I
n
other
wor
d,
f
r
om
th
is
e
quivale
nc
e
c
las
s
,
a
s
e
t
of
a
ll
it
e
ms
e
ts
s
ha
r
ing
the
pr
e
f
ix
c
ould
be
ge
ne
r
a
ted
a
nd
thi
s
s
e
t
f
or
ms
a
s
ub
tr
e
e
o
f
t
he
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
.
W
e
r
e
f
e
r
tr
a
dit
ional
E
c
lat
a
lgo
r
it
h
m
with
E
c
lat
-
ti
ds
e
t
.
I
t
s
tar
ts
with
pr
e
f
ix
{}
a
nd
the
s
e
a
r
c
h
tr
e
e
is
a
c
tual
ly
the
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
.
T
o
divi
de
t
he
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
,
it
picks
the
p
r
e
f
ix
{
}
,
ge
ne
r
a
tes
the
c
or
r
e
s
ponding
e
quivale
nc
e
c
las
s
a
nd
doe
s
f
r
e
que
nt
it
e
ms
e
t
mi
ning
in
the
s
ub
t
r
e
e
of
a
ll
it
e
ms
e
ts
c
ontaining
{
}
,
in
thi
s
s
ub
tr
e
e
it
divi
de
s
f
ur
ther
int
o
tw
o
s
ub
tr
e
e
s
by
picking
the
pr
e
f
ix
{
}
:
the
f
ir
s
t
s
ub
tr
e
e
c
ons
is
ts
of
a
ll
it
e
ms
e
t
c
ontaining
{
}
,
the
other
c
ons
is
ts
of
a
ll
it
e
ms
e
ts
c
ontaining
{
}
but
not
{
}
,
a
nd
thi
s
pr
oc
e
s
s
i
s
r
e
c
ur
s
ive
unti
l
a
ll
it
e
ms
e
ts
in
the
ini
ti
a
l
s
e
a
r
c
h
tr
e
e
a
r
e
vis
it
e
d.
3.
3.
Horiz
on
t
al
vs
.
v
e
r
t
ical
d
a
t
ab
as
e
layou
t
T
he
wor
ks
by
[
15
-
18
]
de
mons
tr
a
te
how
ve
r
ti
c
a
l
da
taba
s
e
layout
ha
s
major
a
dva
ntage
s
ove
r
hor
izonta
l
da
taba
s
e
layout
.
F
i
r
s
tl
y,
c
omput
ing
s
uppor
ts
of
it
e
ms
e
ts
is
much
s
im
pler
a
nd
f
a
s
ter
s
ince
it
invol
ve
s
only
int
e
r
s
e
c
ti
ons
of
ti
ds
a
nd
the
number
of
ti
ds
a
utom
a
ti
c
a
ll
y
indi
c
a
tes
the
s
uppor
t.
I
n
c
ontr
a
s
t,
a
c
ompl
e
x
ha
s
h
-
tr
e
e
da
ta
s
tr
uc
tur
e
s
[
3]
a
nd
f
unc
ti
ons
a
r
e
r
e
quir
e
d
f
or
ho
r
izonta
l
layout.
S
e
c
ondly,
a
n
a
utom
a
ti
c
“
r
e
duc
ti
on”
of
the
da
taba
s
e
be
f
or
e
e
a
c
h
s
c
a
n
whe
r
e
only
thos
e
r
e
leva
nt
it
e
ms
e
ts
to
the
f
oll
owing
s
c
a
n
of
the
mi
ning
pr
oc
e
s
s
a
r
e
a
c
c
e
s
s
e
d
f
r
om
di
s
k.
I
n
ve
r
ti
c
a
l
layout
,
e
a
c
h
it
e
m
in
the
it
e
m
ba
s
e
is
r
e
pr
e
s
e
nted
a
s
:
{
,
(
)
}
a
nd
the
ini
ti
a
l
t
r
a
ns
a
c
ti
on
da
taba
s
e
c
ons
is
ts
of
a
ll
it
e
ms
in
the
it
e
m
ba
s
e
.
F
or
both
layouts
,
it
is
pos
s
ibl
e
to
us
e
the
bit
f
or
mat
to
e
nc
ode
ti
ds
a
nd
a
ls
o
a
c
ombi
na
ti
on
of
both
layouts
c
a
n
be
us
e
d
.
F
igur
e
1
il
lus
tr
a
tes
hor
izont
a
l
a
nd
ve
r
t
ica
l
layout
of
da
ta
r
e
pr
e
s
e
ntation.
F
oll
owing
the
de
pth
-
f
ir
s
t
-
s
e
a
r
c
h
of
pr
e
f
ix
{
}
a
s
s
hown
in
F
igu
r
e
2
,
a
n
e
quivale
nc
e
c
las
s
with
it
e
ms
e
ts
{
,
,
,
}
will
be
ge
ne
r
a
ted,
whic
h
a
r
e
a
ll
2
−
c
ontaining
{
}
.
I
n
thi
s
s
ub
tr
e
e
,
with
the
pr
e
f
ix
{a
b}
,
a
n
e
quivale
nc
e
c
las
s
with
3
−
will
be
{
,
,
}
.
I
t
c
a
n
be
s
e
e
n
that
e
a
c
h
node
in
the
tr
e
e
is
a
pr
e
f
ix
of
a
n
e
quivale
nc
e
c
las
s
with
it
e
ms
e
ts
r
ight
be
low
it
.
He
r
e
,
E
c
lat
doe
s
not
f
ull
y
e
xploi
t
the
downw
a
r
d
c
los
ur
e
pr
ope
r
ty
be
c
a
us
e
of
it
s
de
pth
-
f
ir
s
t
s
e
a
r
c
h.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
i
-
E
c
lat
:
pe
r
for
manc
e
e
nhanc
e
me
nt
of
E
c
lat
v
ia
incr
e
me
ntal
appr
oac
h…
(
W
an
A
e
z
w
ani
W
an
A
bu
B
ak
ar
)
565
F
igur
e
1
.
Hor
izonta
l
a
nd
ve
r
ti
c
a
l
layout
F
igur
e
2.
P
s
e
udoc
ode
f
or
E
c
lat
a
lgor
it
hm
4.
E
CL
AT
VA
R
I
AN
T
S
4.
1.
d
E
c
lat
algorit
h
m
(
E
c
lat
-
d
if
f
s
e
t
)
T
he
dE
c
lat
(
di
f
f
e
r
e
nt
s
e
t
or
dif
f
s
e
t)
r
e
pr
e
s
e
nts
a
n
it
e
ms
e
t
by
ti
ds
that
a
ppe
a
r
in
the
ti
ds
e
t
of
it
s
p
r
e
f
ix
but
do
not
a
ppe
a
r
in
it
s
ti
ds
e
ts
[5
]
.
I
n
other
wor
ds
,
di
f
f
s
e
t
is
the
dif
f
e
r
e
nc
e
be
twe
e
n
two
(
2)
ti
ds
e
ts
(
i.
e
.
ti
ds
e
t
of
the
it
e
ms
e
ts
a
nd
it
s
p
r
e
f
ix)
.
Us
ing
d
i
f
f
s
e
t,
the
c
a
r
dinalit
y
of
s
e
ts
r
e
pr
e
s
e
nti
ng
it
e
ms
e
ts
is
r
e
duc
e
d
s
igni
f
ica
ntl
y
a
nd
thi
s
r
e
s
ult
s
in
f
a
s
ter
int
e
r
s
e
c
ti
on
a
nd
les
s
memor
y
us
a
ge
.
An
e
quivale
nc
e
c
las
s
with
pr
e
f
ix
is
c
ons
ider
e
d
to
c
ontain
the
it
e
ms
e
ts
a
nd
.
L
e
t
(
)
de
notes
the
ti
ds
e
t
o
f
a
nd
d
(
)
)
de
notes
the
di
f
f
s
e
t
o
f
.
W
he
n
us
ing
ti
ds
e
t
f
or
mat,
the
(
)
a
nd
(
)
a
r
e
a
va
il
a
ble
i
n
the
e
quivale
nc
e
c
las
s
a
nd
to
obtain
(
)
,
the
c
a
r
dinalit
y
of
(
)
∩
(
)
=
(
)
c
a
n
be
c
he
c
ke
d
[
6]
.
4.
2.
S
or
t
d
if
f
s
e
t
algorit
h
m
(
E
c
lat
-
s
or
t
d
if
f
s
e
t
)
T
he
s
or
ti
ng
of
di
f
f
s
e
t
(
s
or
td
if
f
s
e
t)
is
to
e
n
ha
nc
e
d
E
c
lat
du
r
ing
s
witching
c
ondit
ion
[
7]
.
W
he
n
s
witching
pr
oc
e
s
s
take
s
plac
e
,
ther
e
e
xis
t
ti
ds
e
ts
whic
h
do
not
s
a
ti
s
f
y
the
s
witching
c
ondit
ion,
thus
thes
e
ti
ds
e
ts
r
e
main
a
s
ti
ds
e
ts
ins
tea
d
of
dif
f
s
e
t
f
or
mat
.
T
he
s
it
ua
ti
on
r
e
s
ult
s
in
both
ti
ds
e
ts
a
nd
dif
f
s
e
ts
f
or
mat
of
it
e
ms
e
ts
in
pa
r
ti
c
ular
e
quivale
nc
e
c
las
s
a
nd
the
ne
xt
int
e
r
s
e
c
ti
on
pr
oc
e
s
s
will
invol
ve
both
f
or
mats
.
e
c
lat
a
lgor
it
hm
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
562
-
570
566
4.
3.
P
os
t
d
if
f
s
e
t
algorit
h
m
(
E
c
lat
-
p
os
t
d
i
f
f
s
e
t
)
A
s
tudy
in
[
19
,
20]
int
r
oduc
e
s
P
os
tdi
f
f
s
e
t
a
lgo
r
it
hm
that
mi
ne
s
in
ti
ds
e
ts
f
or
mat
f
o
r
the
f
ir
s
t
leve
l
while
mi
ning
in
di
f
f
s
e
t
f
o
r
mat
in
the
ne
xt
leve
l
onwa
r
ds
.
T
he
pe
r
f
or
manc
e
(
in
e
xe
c
uti
on
ti
me)
o
f
mi
ning
it
e
ms
e
ts
a
r
e
c
ompar
e
d
be
twe
e
n
tr
a
dit
ional
E
c
lat
(
ti
ds
e
t)
[
4]
,
d
E
c
lat
[
5
]
,
s
or
tdi
f
f
s
e
t
[
7]
a
nd
p
os
tdi
f
f
s
e
t,
a
nd
ye
t,
P
os
tdi
f
f
s
e
t
tur
ns
to
be
the
thi
r
d
a
f
ter
d
E
c
lat
a
nd
s
or
tdi
f
f
s
e
t
.
5.
T
HE
I
NC
RE
M
E
NT
A
L
AP
P
ROAC
H
5.
1.
I
n
c
r
e
m
e
n
t
al
b
as
e
d
on
a
p
r
iori
an
d
F
P
-
gr
ow
t
h
T
o
im
pr
ove
pe
r
f
or
manc
e
a
nd
a
c
c
ur
a
c
y
of
it
e
ms
e
t
mi
ning,
r
e
c
e
nt
r
e
s
e
a
r
c
he
s
in
[
21
-
25]
f
oc
us
towa
r
ds
pa
r
a
ll
e
l
a
nd
incr
e
men
tal
mi
ning
a
ppr
oa
c
he
s
.
I
nc
r
e
menta
l
mi
ning
in
a
dyna
mi
c
d
a
taba
s
e
is
e
s
tablis
he
d
with
r
e
ga
r
ds
to
the
it
e
ms
e
ts
or
r
e
c
or
ds
of
t
r
a
ns
a
c
ti
on.
I
n
c
r
e
menta
l
in
it
e
ms
e
ts
mea
ns
a
n
a
ddit
ional
ne
w
it
e
ms
be
ing
a
dde
d
or
de
lete
d
to
the
e
xis
ti
ng
it
e
ms
e
ts
in
da
ta
ba
s
e
whe
r
e
a
s
incr
e
menta
l
in
r
e
c
or
ds
of
tr
a
ns
a
c
ti
on
mea
ns
the
a
ddit
io
na
l
tr
a
ns
a
c
ti
ons
to
the
e
xis
ti
ng
da
taba
s
e
tr
a
ns
a
c
ti
on.
T
he
a
lgor
it
h
m
f
o
r
mi
ning
we
ight
e
d
maximal
f
r
e
que
nt
it
e
ms
e
ts
f
r
om
inc
r
e
menta
l
da
taba
s
e
s
is
i
ntr
oduc
e
d
[
26
]
.
B
y
s
c
a
nning
a
given
incr
e
menta
l
da
taba
s
e
only
onc
e
,
the
pr
opos
e
d
a
lgor
it
hm
e
xtr
a
c
ts
a
s
maller
number
of
i
mpor
tant
it
e
ms
e
ts
a
nd
pr
ovi
de
mor
e
mea
ningf
ul
pa
tt
e
r
n
r
e
s
ult
s
r
e
f
lec
ti
ng
c
ha
r
a
c
ter
is
ti
c
s
of
g
iven
incr
e
menta
l
da
taba
s
e
s
a
nd
thr
e
s
hold
s
e
tt
ings
.
A
thr
e
e
-
wa
y
de
c
is
i
on
upda
te
pa
tt
e
r
n
a
ppr
oa
c
h
(
T
DU
P
)
is
int
r
oduc
e
d
[2
7]
.
T
he
a
ppr
oa
c
h
o
f
f
e
r
s
f
or
two
s
uppor
t
-
ba
s
e
d
mea
s
ur
e
s
a
nd
s
ync
hr
oniza
ti
on
mec
ha
nis
m
is
pe
r
iodi
c
a
ll
y
tr
igger
e
d
to
r
e
c
omput
e
the
it
e
ms
e
ts
of
f
li
ne
a
nd
r
e
s
ult
s
indi
c
a
te
the
pr
opos
e
d
a
ppr
oa
c
h
e
f
f
icie
nt
a
nd
r
e
l
iable
.
T
he
r
e
s
e
a
r
c
h
in
[
2
8
]
p
r
opos
e
s
incr
e
menta
l
pa
r
a
ll
e
l
a
pr
io
r
i
-
ba
s
e
d
a
lgor
it
hm
on
s
pa
r
k
(
incr
e
menta
l
F
I
M
)
whe
r
e
it
upda
tes
f
r
e
que
nt
it
e
ms
e
t
ba
s
e
d
on
pr
e
vious
f
r
e
que
nt
it
e
ms
e
t
ins
tea
d
o
f
r
e
c
o
mput
iong
the
whole
da
tas
e
ts
f
r
om
s
c
r
a
tch.
T
he
r
e
s
ult
s
hows
the
a
lgor
it
hm
im
p
r
ove
s
mi
ning
pe
r
f
o
r
manc
e
s
igni
f
ica
ntl
y.
T
he
I
F
I
N
a
lgor
it
hm
(
a
pr
e
f
ix
tr
e
e
s
tr
uc
tur
e
I
P
P
C
-
T
r
e
e
ba
s
e
d
on
F
P
-
Gr
owth
is
de
ve
loped
[
29]
.
T
he
a
lgor
it
hm
maintains
a
loca
l
a
nd
c
ha
nge
a
ble
or
de
r
o
f
it
e
m
in
a
pa
th
node
s
f
r
om
the
r
oot
to
a
lea
ve
s
,
whic
h
doe
s
not
wa
s
te
c
omput
a
ti
ona
l
ove
r
he
a
d
f
or
the
p
r
e
vious
ly
pr
oc
e
s
s
e
d
pa
r
t
of
da
ta
whe
n
a
ne
w
it
e
ms
e
t
is
a
dde
d
or
the
s
uppor
t
th
r
e
s
hold
is
c
ha
nge
d
a
nd
the
r
e
s
ult
de
picts
a
n
e
f
f
icie
nt
ti
me
a
nd
memo
r
y
c
ons
umpt
ion
in
mi
ni
ng.
5.
2.
I
n
c
r
e
m
e
n
t
al
b
as
e
d
on
e
c
lat
(
i
-
E
c
lat
)
Our
a
ppr
oa
c
h
is
to
incr
e
ment
the
mi
ning
of
it
e
ms
e
ts
f
r
om
E
c
lat
a
s
our
ba
s
e
d
model.
W
e
a
dd
with
two
(
2)
ba
s
ic
de
f
ini
ti
ons
of
incr
e
menta
l
mi
ning
s
uc
h
a
s
f
oll
ows
:
De
f
ini
ti
on
5.
(
I
nc
r
e
menta
l
Da
taba
s
e
)
:
Give
n
a
s
e
que
nc
e
of
tr
a
ns
a
c
ti
on,
T
,
e
a
c
h
o
f
whic
h
is
c
a
ll
e
d
it
e
ms
e
t.
F
o
r
a
da
taba
s
e
D
a
nd
a
s
e
que
nc
e
α
,
1the
s
uppor
t
of
α
in
D
is
de
noted
by
s
u
p
p
o
r
t
D
(
α
)
is
the
f
r
e
que
nc
y
of
it
e
ms
in
D
.
S
uppos
e
that
ne
w
da
ta,
δ
is
to
be
a
dde
d
to
da
taba
s
e
D
.
T
he
n
D
is
s
a
id
to
be
or
igi
na
l
da
taba
s
e
a
nd
δ
is
the
incr
e
menta
l
da
tab
a
s
e
.
T
he
upda
ted
da
taba
s
e
is
de
noted
by
D
+
δ
.
De
f
ini
ti
on
6.
(
I
nc
r
e
menta
l
R
e
c
or
ds
a
nd
I
tems
e
ts
Dis
c
ove
r
y
P
r
oblem)
:
Give
n
an
or
igi
na
l
da
taba
s
e
D
a
nd
a
ne
w
incr
e
ment
to
the
D
whic
h
is
δ
,
f
ind
a
ll
f
r
e
que
nt
it
e
ms
e
ts
in
da
taba
s
e
(
D
+
δ
)
with
m
ini
mum
pos
s
ibl
e
r
e
c
omput
a
ti
on
a
nd
I
/O
ove
r
he
a
ds
.
F
o
r
e
a
c
h
k
≥
1
,
F
k
de
notes
the
c
oll
e
c
ti
on
of
f
r
e
que
nt
it
e
ms
e
ts
of
length
k
in
the
upda
ted
da
taba
s
e
(
D
+
δ
)
.
A
phys
ica
l
de
s
ign
of
i
-
E
c
lat
is
diag
r
a
mm
e
d
in
F
igu
r
e
3.
F
r
om
or
igi
na
l
da
taba
s
e
,
a
ll
f
r
e
que
nt
it
e
ms
a
r
e
pa
s
s
e
d
to
the
f
ir
s
t
pr
uning
pr
oc
e
s
s
,
ge
taw
a
y
G1
.
G1
is
s
e
t
with
the
va
lue
of
mi
n_s
uppor
t
thr
e
s
hold
pr
ior
to
ge
ne
r
a
ti
ng
the
li
s
t
of
c
a
ndidate
s
.
T
o
s
e
t
G1,
tot
a
l
tr
a
ns
a
c
ti
on
r
e
c
or
ds
a
r
e
s
c
a
nne
d
to
be
mul
ti
pl
ied
with
the
pe
r
c
e
ntage
of
us
e
r
-
s
pe
c
if
ied
mi
n_s
uppor
t
va
l
ue
.
Onc
e
the
va
lue
is
obtaine
d,
only
c
a
ndidate
of
f
r
e
que
nt
it
e
ms
e
ts
that
pa
s
s
e
d
the
G1
va
lue
is
pr
oc
e
s
s
e
d
e
i
ther
thr
ough
E
c
lat
-
ti
ds
e
t,
E
c
lat
-
dif
f
s
e
t,
E
c
lat
-
s
or
td
if
f
s
e
t
or
E
c
lat
-
pos
tdi
f
f
s
e
t
a
lgor
it
hms
in
E
c
lat
e
ngine.
S
e
c
on
d
pr
uning
pr
oc
e
s
s
,
ge
taw
a
y
2A
o
r
G2A
whe
r
e
da
ta
i
s
wr
it
ten
to
text
f
il
e
in
i
-
E
c
lat
e
ngine.
B
y
wr
it
ing
to
text
f
i
le,
c
a
ndidate
it
e
ms
e
ts
a
r
e
dir
e
c
ted
to
ha
r
d
dis
k
s
tor
a
g
e
,
s
o
that
the
r
e
s
our
c
e
of
memor
y
s
tor
a
ge
is
a
utom
a
ti
c
a
ll
y
r
e
duc
e
d
to
e
na
ble
the
pr
oc
e
s
s
ing
a
nd
e
xe
c
uti
ng
of
f
ull
da
tas
e
ts
.
T
his
is
due
to
maximum
memor
y
c
ons
umpt
ion
in
ha
ndli
ng
too
many
c
a
ndidate
it
e
ms
e
ts
dur
ing
int
e
r
s
e
c
ti
ng
pr
oc
e
s
s
.
I
nc
r
e
menta
l
a
ppr
oa
c
h
is
done
in
t
r
a
ns
a
c
ti
on
(
a
ddin
g
r
ow)
o
r
in
nu
mber
of
it
e
ms
e
ts
(
a
dding
c
olum
n)
.
T
he
a
dva
ntage
of
a
n
incr
e
menta
l
s
tor
a
ge
s
tr
uc
tur
e
is
that
e
a
c
h
c
a
ndidate
it
e
ms
e
ts
in
the
s
e
a
r
c
h
s
pa
c
e
ha
s
it
s
c
ounter
pa
r
t
in
the
da
taba
s
e
,
s
uc
h
that
it
s
s
uppor
t
c
a
n
be
c
omput
e
d
by
a
f
e
w
s
im
ple
da
taba
s
e
op
e
r
a
ti
o
ns
,
r
a
ther
than
a
f
ull
s
c
a
n
of
a
da
taba
s
e
.
T
he
s
uppor
t
o
f
a
1
-
it
e
ms
e
t,
is
s
im
ply
the
s
ize
of
c
olum
n
in
the
da
taba
s
e
.
S
o
in
the
f
ir
s
t
pa
s
s
of
the
la
r
ge
s
e
t
d
is
c
ove
r
y
a
lgor
i
thm
,
it
only
ha
s
to
s
e
lec
t
1
−
whos
e
c
olum
ns
ha
ve
s
ize
a
bove
s
upp
or
t
thr
e
s
hold
.
T
he
s
uppor
t
o
f
2
−
,
is
the
number
of
t
r
a
ns
a
c
ti
ons
that
c
ontain
both
a
nd
.
S
ince
the
ti
ds
f
o
r
a
nd
a
r
e
s
tor
e
d
in
s
e
pa
r
a
te
c
olum
n
in
da
taba
s
e
,
then
how
many
ti
ds
would
a
ppe
a
r
in
both
a
nd
?
T
hus
the
c
omput
a
ti
on
is
the
int
e
r
s
e
c
ti
on
be
twe
e
n
a
nd
i.
e
.
∩
.
B
ut
the
p
r
oblem
a
r
is
e
s
e
s
pe
c
ially
in
the
s
e
c
ond
pa
s
s
whe
r
e
many
c
a
ndidate
s
a
r
e
ge
ne
r
a
ted
with
only
ve
r
y
f
e
w
pr
ove
to
be
lar
ge
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
i
-
E
c
lat
:
pe
r
for
manc
e
e
nhanc
e
me
nt
of
E
c
lat
v
ia
incr
e
me
ntal
appr
oac
h…
(
W
an
A
e
z
w
ani
W
an
A
bu
B
ak
ar
)
567
F
or
e
xa
mpl
e
,
the
r
e
a
r
e
600
it
e
ms
e
ts
out
of
1000
it
e
ms
e
ts
in
the
f
ir
s
t
pa
s
s
a
r
e
lar
ge
.
T
he
n,
in
the
s
e
c
ond
pa
s
s
,
the
it
e
ms
e
ts
will
be
600
2
/
2
=
1
8
0
0
0
0
c
a
ndidate
s
of
whic
h
only
44
!
a
r
e
a
c
tually
lar
ge
.
T
hus
,
to
f
ind
thes
e
,
ther
e
a
r
e
180000
int
e
r
s
e
c
ti
ons
that
c
ons
umes
ove
r
99%
of
tot
a
l
da
taba
s
e
pr
oc
e
s
s
ing
ti
me
[
30
]
.
T
he
ter
mi
nology
min_s
upp
us
e
d
in
the
ps
e
udoc
ode
is
to
r
e
f
e
r
to
min_s
uppor
t
thr
e
s
hold
va
lue
that
is
de
ter
mi
ne
d
in
ter
ms
o
f
pe
r
c
e
ntage
whe
r
e
the
us
e
r
-
s
pe
c
if
ied
mi
n_
s
upp
va
lue
will
be
divi
de
d
by
100
a
nd
mul
ti
ply
with
tot
a
l
r
ows
(
r
e
c
or
ds
)
of
e
a
c
h
da
tas
e
ts
.
Ne
xt
in
e
a
c
h
loop,
s
tar
ti
ng
with
the
f
ir
s
t
loop,
if
the
s
uppor
t
i
s
gr
e
a
ter
than
or
e
qua
l
(
>
=
)
to
mi
n_s
upp,
then
,
ge
tt
ing
the
ti
ds
e
t
int
e
r
s
e
c
ti
on
a
nd
dif
fs
e
t
int
e
r
s
e
c
ti
on
in
i
-
E
c
lat
-
ti
d
s
e
t
a
nd
i
-
E
c
la
t
-
dif
fs
e
t
r
e
s
pe
c
ti
ve
ly
be
twe
e
n
ℎ
c
olum
n
a
nd
+
1
ℎ
c
olum
n
a
nd
s
a
ve
to
db.
I
n
i
-
E
c
lat
-
s
or
tdi
ff
s
e
t
,
it
e
ms
e
ts
a
r
e
f
ir
s
t
s
or
ted
in
de
s
c
e
nding
or
de
r
d
e
pe
nding
upon
the
highes
t
to
lowe
s
t
va
lue
of
it
e
ms
e
t’
s
e
quivale
nc
e
c
las
s
.
T
he
n
the
di
f
f
s
e
t
va
lue
be
twe
e
n
ℎ
c
olum
n
a
nd
+
1
ℎ
c
olum
n
will
be
e
nc
ounter
e
d
a
nd
s
a
ve
to
db
while
f
o
r
i
-
E
c
lat
-
pos
tdi
ff
s
e
t
,
the
f
ir
s
t
leve
l
of
loopi
ng
is
ba
s
e
d
on
ti
ds
e
ts
pr
oc
e
s
s
,
f
o
ll
ows
by
the
s
e
c
ond
leve
l
onwa
r
ds
o
f
loopi
ng
a
r
e
ge
tt
ing
th
e
r
e
s
ult
of
dif
f
s
e
t
be
twe
e
n
ℎ
c
olum
n
a
nd
+
1
ℎ
c
olum
n
be
f
or
e
s
a
ving
to
db
.
F
igur
e
3.
P
hys
ica
l
de
s
ign
of
i
-
E
c
lat
e
ngine
6.
E
XP
E
RI
M
E
NT
AT
I
ON
All
e
xpe
r
im
e
nts
a
r
e
pe
r
f
or
med
on
a
De
ll
N5050,
I
ntel®P
e
nti
um
®C
P
U
B
960@2.
20
GH
z
with
8
GB
R
AM
in
a
W
in
7
64
-
bit
platf
or
m
.
T
he
s
of
twa
r
e
s
pe
c
if
ica
ti
on
f
or
a
lgo
r
it
hm
de
ve
lopm
e
nt
is
de
ploy
e
d
us
ing
M
yS
QL
ve
r
s
ion
5.
6.
20,
Apa
c
he
/2.
4.
10
(
W
in32
)
Ope
nS
S
L
/1.
0.
1i
P
HP/
5.
5
.
15
f
or
ou
r
we
b
s
e
r
ve
r
,
php
a
s
a
pr
ogr
a
mm
ing
langua
ge
.
T
he
r
e
tr
ieva
l
of
be
nc
hmar
k
da
tas
e
ts
a
r
e
f
r
om
(
Goe
thals
,
2003)
in
a
*.
dat
f
il
e
f
or
mat.
T
he
two
(
2
)
c
a
tegor
y
of
da
tas
e
ts
a
r
e
de
ns
e
(
i.
e
.
a
di
mens
ion
with
a
high
pr
oba
bil
it
y
that
one
or
mor
e
d
a
ta
point
s
is
oc
c
upied
in
e
ve
r
y
c
ombi
na
ti
on
of
dim
e
ns
ions
)
a
nd
s
pa
r
s
e
(
i.
e
.
a
dim
e
ns
ion
with
a
l
ow
pe
r
c
e
ntage
of
a
va
il
a
ble
da
ta
pos
it
ions
f
il
led)
.
T
he
ove
r
a
ll
c
ha
r
a
c
ter
is
ti
c
s
of
be
nc
hmar
k
da
tas
e
ts
is
tabula
ted
in
T
a
ble
1.
T
a
ble
1.
Da
taba
s
e
c
ha
r
a
c
ter
is
ti
c
s
D
a
ta
ba
s
e
#S
iz
e
(
K
B
)
# L
e
ngt
h (
a
tt
r
ib
ut
e
)
#I
te
m
#R
e
c
or
ds
(
tr
a
ns
a
c
ti
on)
C
a
te
gor
y
C
he
s
s
334
37
75
3196
D
e
ns
e
M
us
hr
oom
557
23
119
8124
D
e
ns
e
P
ums
b_s
ta
r
1130
48
2088
49046
D
e
ns
e
R
e
ta
il
9490
45
16469
88162
S
pa
r
s
e
T
10I
4D
100K
5630
26
1000
100000
S
pa
r
s
e
7.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
he
pe
r
f
or
manc
e
of
f
ive
(
5)
de
ns
e
a
nd
s
pa
r
s
e
da
tas
e
t
s
is
mea
s
ur
e
d
ba
s
e
d
on
(
1
)
given.
T
he
e
xa
mpl
e
of
pe
r
c
e
ntage
of
r
e
duc
ti
on
r
a
ti
o
o
f
ti
ds
in
a
s
c
ompar
e
d
to
ti
ds
in
is
c
a
lcula
ted
ba
s
e
d
o
n
(
1
)
that
de
ter
mi
ne
s
the
outper
f
or
m
pe
r
c
e
ntage
of
.
(
)
−
(
)
100
%
(
1)
W
e
r
e
ve
a
ls
the
e
xpe
r
im
e
ntation
with
only
taking
50
%
mi
n_s
upp
thr
e
s
hold
va
lue
that
we
tes
t
f
or
ti
ds
e
t,
dif
f
s
e
t,
s
or
td
if
f
s
e
t
a
nd
pos
tdi
f
f
s
e
t
a
lgo
r
it
hms
.
F
ig
ur
e
4
p
lot
s
the
g
r
a
ph
o
f
f
ull
c
he
s
s
da
tas
e
t
r
unning
in
E
c
lat
a
lgor
it
hm
a
n
d
de
ve
loped
i
-
E
c
lat
a
lgor
it
hm.
T
he
i
-
E
c
lat
outper
f
o
r
ms
in
ti
ds
e
t,
dif
f
s
e
t
a
nd
pos
tdi
f
f
s
e
t
with
3%
,
32%
a
nd
54%
r
e
s
pe
c
ti
ve
ly
les
s
e
r
in
e
xe
c
uti
on
ti
me,
while
E
c
lat
outper
f
or
ms
with
48
%
in
s
or
tdi
f
f
s
e
t
a
lgor
it
hm.
Dif
f
s
e
t
s
hows
a
tr
e
mendous
r
e
s
ult
in
e
xe
c
uti
on
ti
me
with
the
lea
s
t
in
s
e
c
onds
.
T
he
mus
hr
oom
da
tas
e
t
pr
ojec
ts
D
a
ta
ba
s
e
L
i
s
t
o
f
C
a
n
d
i
d
a
t
e
s
L
is
t
of
F
r
e
que
nt
I
te
ms
Fr
e
q
u
e
n
t
AR
G
1
i
-
E
c
la
t
E
n
gi
n
e
ti
ds
e
t
di
f
f
s
e
t
s
or
td
if
f
s
e
t
G
2
A
pos
td
if
f
s
e
t
P
r
oc
e
s
s
F
r
e
que
nt
I
te
ms
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
562
-
570
568
a
bout
a
dif
f
e
r
e
nt
t
r
e
nd
in
pa
tt
e
r
n
be
twe
e
n
E
c
lat
a
nd
i
-
E
c
lat
e
ngine
a
s
c
ompar
e
d
to
c
he
s
s
a
s
il
lus
tr
a
ted
in
F
igur
e
5.
i
-
E
c
lat
outper
f
or
ms
in
ti
ds
e
t,
di
f
f
s
e
t
a
nd
p
os
tdi
f
f
s
e
t
with
74%
,
71
%
a
nd
63
%
be
tt
e
r
.
B
ut
in
s
o
r
tdi
f
f
s
e
t,
it
tur
ns
to
E
c
lat
whic
h
pe
r
f
o
r
ms
be
tt
e
r
with
81
in
p
e
r
c
e
ntage
.
T
he
thi
r
d
de
ns
e
da
tas
e
t,
pums
b_s
tar
s
how
s
only
a
s
mall
dif
f
e
r
e
nc
e
of
pe
r
f
or
manc
e
be
twe
e
n
i
-
E
c
lat
a
nd
E
c
lat
e
ngines
.
F
igur
e
6
plot
s
the
gr
a
phs
of
f
ul
l
pums
b_s
tar
s
uc
h
that
i
-
E
c
lat
outper
f
or
ms
only
a
t
a
s
mall
r
a
ti
o
whe
r
e
only
10%
is
r
e
c
or
de
d
in
d
if
f
s
e
t.
T
his
f
oll
ows
by
s
or
tdi
f
f
s
e
t,
pos
tdi
f
f
s
e
t
a
nd
ti
ds
e
t
wi
th
only
6%
,
3%
a
nd
1%
be
tt
e
r
pe
r
f
or
manc
e
.
T
he
r
e
s
ult
s
of
f
ir
s
t
in
s
pa
r
s
e
c
a
tegor
y
whic
h
is
r
e
tail
da
tas
e
t
c
lea
r
ly
de
picts
in
F
igur
e
7
.
T
he
pe
r
c
e
ntage
r
e
c
or
de
d
in
a
ll
a
lgor
i
th
ms
of
r
e
tail
da
tas
e
t
s
hows
a
d
if
f
e
r
e
nt
t
r
e
nd
a
s
c
ompar
e
d
to
other
de
ns
e
da
tas
e
ts
whe
r
e
E
c
lat
e
ngine
ou
tper
f
or
ms
in
mos
t
o
f
r
e
tail
da
tas
e
t
in
s
mall
s
e
gments
e
xc
e
pt
in
f
ull
r
e
tail
da
tas
e
t.
Ove
r
a
ll
pe
r
f
or
manc
e
e
va
luation
o
f
r
e
tail
s
hows
that
E
c
lat
wins
with
only
5%
,
16
%
,
9
%
a
nd
7%
be
tt
e
r
pe
r
f
or
manc
e
in
ti
ds
e
t,
d
if
f
s
e
t,
s
or
tdi
f
f
s
e
t
a
nd
pos
tdi
f
f
s
e
t
r
e
s
pe
c
ti
ve
ly.
T
he
T
10I
4D100K
d
a
t
a
s
e
t
is
the
las
t
in
s
pa
r
s
e
c
a
tegor
y
in
thi
s
e
xpe
r
im
e
nt
a
ti
on.
Unl
ike
in
r
e
tail,
i
-
E
c
lat
outper
f
o
r
ms
in
mos
t
of
the
a
lgor
it
hms
e
it
he
r
in
f
ull
da
tas
e
t
or
in
s
mall
inc
r
e
ment
of
r
e
c
or
ds
a
nd
it
e
ms
e
ts
.
F
igur
e
8
de
picts
t
he
gr
a
ph
whe
r
e
i
-
E
c
lat
e
ngine
s
e
e
ms
to
ou
tper
f
o
r
m
with
14
%
be
tt
e
r
in
ti
ds
e
t,
17%
in
di
f
f
s
e
t,
on
ly
6%
in
s
or
td
if
f
s
e
t
while
the
las
t
pos
tdi
f
f
s
e
t
is
r
e
c
or
de
d
a
s
31%
be
tt
e
r
in
ti
m
e
e
xe
c
uti
on.
T
o
s
umm
a
r
ize
,
the
pe
r
f
or
manc
e
o
f
e
i
ther
E
c
lat
e
ngi
ne
that
is
de
ve
loped
by
f
lus
hing
the
c
a
c
he
memor
y
(
us
ing
me
mor
y
c
a
pa
c
it
y)
or
i
-
E
c
lat
e
ngine
that
is
de
ve
loped
by
wr
it
ing
to
text
f
il
e
(
us
ing
dis
k
s
tor
a
ge
c
a
pa
c
it
y)
,
both
e
ngines
c
onf
ir
ms
the
be
s
t
a
lgor
it
hm
in
ve
r
ti
c
a
l
mi
ning
of
f
r
e
que
nt
it
e
ms
e
ts
be
longs
to
dif
f
s
e
t
a
lgor
it
hm
[
5
]
,
f
oll
ows
by
s
or
tdi
f
f
s
e
t
[
7]
,
ne
xt
i
s
the
ne
w
de
ve
loped
pos
tdi
f
f
s
e
t
a
lgor
it
hm
[
19
]
by
the
r
e
s
e
a
r
c
he
r
a
nd
the
las
t
r
a
nking
goe
s
to
ti
ds
e
t
[
4
]
.
F
o
r
t
he
pe
r
f
or
manc
e
o
f
s
or
tdi
f
f
s
e
t,
the
r
e
s
ult
s
s
e
e
m
to
c
ontr
a
dict
with
[
7
]
.
Due
to
M
yS
QL
im
p
leme
ntation,
the
ti
me
t
a
ke
n
f
or
s
or
ti
ng
the
dif
f
s
e
t
of
it
e
ms
e
ts
ha
s
r
e
s
ult
e
d
i
n
longer
ti
me
e
xe
c
uti
on.
T
o
obs
e
r
ve
on
pos
tdi
f
f
s
e
t
a
lgo
r
it
hm,
it
tur
ns
to
ou
tper
f
or
m
in
i
-
E
c
lat
e
ngine
with
63%
in
mus
hr
oom,
31
%
in
T
10I
4D100K
a
nd
only
3%
in
b
oth
pums
b_s
tar
a
nd
r
e
tail.
How
e
ve
r
,
in
c
he
s
s
,
pos
tdi
f
f
s
e
t
i
s
outs
tanding
in
E
c
lat
e
ngine
f
o
r
54%
.
F
o
r
ove
r
a
ll
d
a
tas
e
ts
,
the
r
e
s
ult
s
indi
c
a
te
that
i
-
E
c
lat
outper
f
or
m
E
c
lat
f
or
17%
both
in
c
he
s
s
a
nd
T
10I
4D100K
,
69
%
in
mus
hr
oom,
5
%
a
nd
8%
in
pums
b_s
tar
a
nd
r
e
tail
da
tas
e
ts
.
F
r
om
thes
e
pe
r
c
e
ntage
f
igur
e
s
,
divi
d
ing
wi
th
f
ive
(
5)
d
a
tas
e
ts
,
a
ve
r
a
ge
pe
r
f
or
manc
e
of
i
-
E
c
lat
is
c
onc
lud
e
d
to
be
23.
37%
be
tt
e
r
than
E
c
lat
e
ngine.
F
igur
e
4.
C
he
s
s
in
E
c
lat
e
ngine
v
s
i
-
E
c
lat
e
ngine
F
igur
e
5.
M
us
hr
oom
in
E
c
lat
e
ngine
vs
i
-
E
c
lat
e
ngine
F
igur
e
6
.
P
ums
b_s
tar
in
E
c
lat
e
ngine
vs
i
-
E
c
lat
e
ngine
F
igur
e
7.
R
e
tail
in
E
c
lat
e
ngine
vs
i
-
E
c
lat
e
ngine
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
i
-
E
c
lat
:
pe
r
for
manc
e
e
nhanc
e
me
nt
of
E
c
lat
v
ia
incr
e
me
ntal
appr
oac
h…
(
W
an
A
e
z
w
ani
W
an
A
bu
B
ak
ar
)
569
F
igur
e
8.
T
10
I
4D100K
in
E
c
lat
e
ngine
vs
i
-
E
c
lat
e
ngine
8.
CONC
L
USI
ON
AN
D
F
UT
UR
E
DI
RE
C
T
I
ONS
T
he
r
e
s
e
a
r
c
h
pr
ove
s
that
the
mo
r
e
inc
r
e
ment
in
it
e
ms
e
t
(
c
olum
n)
r
e
s
ult
ing
in
the
mor
e
us
a
ge
o
f
memor
y
a
s
c
ompar
e
d
to
the
inc
r
e
ment
o
f
r
e
c
or
ds
o
f
t
r
a
n
s
a
c
ti
on
a
s
indi
c
a
ted
in
F
igu
r
e
4
to
F
igur
e
8.
T
his
is
due
to
the
incr
e
ment
of
it
e
ms
e
ts
pr
oduc
e
s
the
higher
c
a
r
dinalit
y
of
int
e
r
s
e
c
ti
on
be
twe
e
n
e
a
c
h
it
e
m
that
ne
e
d
s
to
be
c
onduc
ted
in
ve
r
t
ica
l
mi
ning
.
T
ha
t
is
why
the
mu
c
h
higher
e
xe
c
uti
on
ti
me
c
a
n
be
s
e
e
n
in
c
he
s
s
,
pu
ms
b_s
tar
a
nd
r
e
tail
da
tas
e
ts
in
s
pit
e
of
mus
hr
oom
a
nd
T
10I
4D100K
da
tas
e
ts
.
E
it
he
r
E
c
lat
or
i
-
E
c
lat
e
ngine,
the
pe
r
f
or
manc
e
o
f
both
e
ngines
is
a
c
tually
de
pe
nd
upon
the
na
tur
e
o
f
da
tas
e
t
it
s
e
lf
whe
n
tes
ti
ng
in
ti
ds
e
t,
dif
f
s
e
t,
s
or
tdi
f
f
s
e
t
a
nd
pos
tdi
f
f
s
e
t
a
lgor
it
hms
.
Ho
we
ve
r
,
both
e
ngines
c
onf
o
r
ms
that
a
mong
thes
e
f
our
(
4)
a
lgor
it
hms
,
di
f
f
s
e
t
outper
f
or
ms
othe
r
a
lgor
it
hms
by
c
e
r
tain
or
de
r
of
magnitude
in
a
l
l
s
e
lec
ted
da
tas
e
ts
.
F
or
ou
r
ne
xt
dir
e
c
ti
on
in
e
xpe
r
i
menta
ti
on,
we
may
tac
kle
the
is
s
ue
of
inf
r
e
que
nt
pa
tt
e
r
n,
whe
ther
it
c
a
n
f
it
a
nd
pr
ovide
hidden
va
luable
pa
tt
e
r
n
f
o
r
a
na
lys
is
.
AC
KNOWL
E
DGE
M
E
NT
S
S
pe
c
ial
gr
a
ti
tudes
to
R
M
I
C
of
UniS
Z
A
f
or
pr
ovid
i
ng
f
inanc
ial
s
uppor
t
unde
r
UniS
Z
A
int
e
r
na
l
g
r
a
nt
c
ode
(
UN
I
S
Z
A/2018/DP
U/11)
a
nd
a
ll
tea
m
membe
r
s
e
s
pe
c
ially
tea
m
s
f
r
om
UM
T
a
nd
UM
K
f
or
mo
r
a
le
a
nd
tec
hnica
l
s
uppor
t.
RE
F
E
RE
NC
E
S
[1
]
A
g
ra
w
al
R,
Sri
k
a
n
t
R.
,
“
Fas
t
al
g
o
ri
t
h
m
s
fo
r
m
i
n
i
n
g
as
s
o
c
i
at
i
o
n
ru
l
e
s
,
”
P
r
o
c.
2
0
th
i
n
t
.
co
n
f
.
ver
y
l
a
r
g
e
d
a
t
a
b
a
s
e
s
V
LD
B
,
v
o
l
.
1
2
1
5
,
p
p
.
4
8
7
-
4
9
9
,
1
9
9
4
.
[2
]
A
g
ra
w
al
R,
Imi
el
i
ń
s
k
i
T
,
Sw
am
i
A
.
,
“
Mi
n
i
n
g
as
s
o
ci
at
i
o
n
ru
l
es
b
et
w
een
s
et
s
o
f
i
t
ems
i
n
l
arg
e
d
at
a
b
a
s
es
”
.
P
r
o
ceed
i
n
g
s
A
c
m
s
i
g
m
o
d
r
eco
r
d
,
v
o
l
.
2
2
,
n
o
.
2
,
p
p
.
2
0
7
-
216,
1
9
9
3
.
[3
]
H
an
J
,
Pei
J
,
Y
i
n
Y
.
“
Mi
n
i
n
g
freq
u
e
n
t
p
a
t
t
er
n
s
w
i
t
h
o
u
t
can
d
i
d
at
e
g
en
era
t
i
o
n
,”
P
r
o
ceed
i
n
g
s
A
C
M
s
i
g
m
o
d
r
ec
o
r
d
,
v
o
l
.
2
9
,
N
o
.
2
,
p
p
.
1
-
1
2
,
2
0
0
0
.
[4
]
O
g
i
h
ara
Z
.
P.
,
Z
ak
i
M.
J
.
,
Part
h
a
s
arat
h
y
S.
,
O
g
i
h
ar
a
M
.
,
L
i
W
.
,
“
N
ew
al
g
o
r
i
t
h
ms
f
o
r
fas
t
d
i
s
c
o
v
er
y
o
f
a
s
s
o
ci
a
t
i
o
n
ru
l
e
s
,
”
3
rd
In
t
l
.
Co
n
f
.
o
n
Kn
o
w
l
ed
g
e
D
i
s
c
o
ver
y
a
n
d
D
a
t
a
M
i
n
i
n
g
,
1
9
9
7
.
[5
]
Z
ak
i
M.
J
.
,
G
o
u
d
a
K
.
,
“
Fa
s
t
v
ert
i
cal
mi
n
i
n
g
u
s
i
n
g
d
i
ffs
et
s
,
”
P
r
o
c
eed
i
n
g
s
o
f
t
h
e
n
i
n
t
h
A
CM
S
IG
K
D
D
i
n
t
e
r
n
a
t
i
o
n
a
l
co
n
f
er
e
n
ce
o
n
K
n
o
w
l
ed
g
e
d
i
s
co
ve
r
y
a
n
d
d
a
t
a
m
i
n
i
n
g
,
p
p
.
3
2
6
-
3
3
5
,
A
u
g
u
s
t
2
0
0
3
.
[6
]
Sh
en
o
y
P.
,
H
ar
i
t
s
a
J
.
R.
,
Su
d
ar
s
h
a
n
S.
,
Bh
a
l
o
t
i
a
G
.
,
B
aw
a
M.
,
Sh
ah
D
.
,
“
T
u
rb
o
-
ch
ar
g
i
n
g
v
er
t
i
ca
l
mi
n
i
n
g
o
f
l
arg
e
d
at
a
b
as
e
s
,”
A
CM
S
i
g
m
o
d
R
ec
o
r
d
,
v
o
l
.
2
9
,
n
o
.
2
,
p
p
.
2
2
-
3
3
,
May
2
0
0
0
.
[7
]
T
ri
e
u
T
.
A
.
,
K
u
n
i
e
d
a
Y
.
,
“
A
n
i
mp
ro
v
eme
n
t
fo
r
d
E
c
l
at
al
g
o
r
i
t
h
m
,”
P
r
o
cee
d
i
n
g
s
o
f
t
h
e
6
th
In
t
er
n
a
t
i
o
n
a
l
Co
n
f
e
r
en
c
e
o
n
U
b
i
q
u
i
t
o
u
s
In
f
o
r
m
a
t
i
o
n
M
a
n
a
g
e
m
en
t
a
n
d
Co
m
m
u
n
i
ca
t
i
o
n
,
n
o
.
5
4
,
p
p
.
1
-
6,
Feb
r
u
ary
2
0
1
2
[8
]
H
i
p
p
J
.
,
G
ü
n
t
zer
U
.
,
N
ak
h
aei
za
d
e
h
G
.
,
“
A
l
g
o
ri
t
h
m
s
fo
r
as
s
o
ci
a
t
i
o
n
ru
l
e
mi
n
i
n
g
-
a
g
en
eral
s
u
rv
e
y
an
d
co
m
p
ari
s
o
n
,
”
S
IG
K
D
D
exp
l
o
r
a
t
i
o
n
s
,
v
o
l
.
2
,
n
o
.
1
,
p
p
.
58
-
6
4
,
J
u
l
y
2
0
0
0
.
[9
]
H
an
J
.
,
Ch
en
g
H
.
,
X
i
n
D
.
,
Y
an
X
.
,
“
Freq
u
en
t
p
at
t
er
n
mi
n
i
n
g
:
cu
rre
n
t
s
t
at
u
s
an
d
fu
t
u
re
d
i
rect
i
o
n
s
,
”
D
a
t
a
m
i
n
i
n
g
a
n
d
kn
o
w
l
ed
g
e
d
i
s
co
ve
r
y
,
v
o
l
.
15
,
n
o
.
1
,
p
p
.
5
5
-
8
6
,
J
u
l
y
2
0
0
7
.
[1
0
]
Bo
rg
e
l
t
C.
,
“
E
ffi
ci
e
n
t
i
m
p
l
e
me
n
t
a
t
i
o
n
s
o
f
ap
r
i
o
r
i
an
d
éc
l
at
,
”
F
IM
I’
0
3
:
P
r
o
ceed
i
n
g
s
o
f
t
h
e
IE
E
E
ICD
M
wo
r
ks
h
o
p
o
n
f
r
e
q
u
e
n
t
i
t
em
s
et
m
i
n
i
n
g
i
m
p
l
em
e
n
t
a
t
i
o
n
s
,
D
ecem
b
er
2
0
0
3
.
[1
1
]
Sch
mi
d
t
-
T
h
i
eme
L
.
,
“A
l
g
o
r
i
t
h
mi
c
Feat
u
res
o
f
E
cl
at
,
”
F
I
M
I
,
N
o
v
emb
er
2
0
0
4
.
[1
2
]
M.
J
.
Z
ak
i
,
"
Sc
al
ab
l
e
al
g
o
ri
t
h
m
s
fo
r
as
s
o
ci
a
t
i
o
n
mi
n
i
n
g
,
"
IE
E
E
Tr
a
n
s
a
ct
i
o
n
s
o
n
Kn
o
w
l
ed
g
e
a
n
d
D
a
t
a
E
n
g
i
n
ee
r
i
n
g
,
v
o
l
.
1
2
,
n
o
.
3
,
p
p
.
3
7
2
-
3
9
0
,
May
-
J
u
n
e
2
0
0
0
.
[1
3
]
Y
u
X
.
,
W
an
g
H
.
,
“
Imp
r
o
v
eme
n
t
o
f
E
c
l
at
a
l
g
o
ri
t
h
m
b
as
e
d
o
n
s
u
p
p
o
r
t
i
n
freq
u
en
t
i
t
ems
e
t
mi
n
i
n
g
,
”
Jo
u
r
n
a
l
o
f
Co
m
p
u
t
er
s
,
v
o
l
.
9
,
n
o
.
9
,
p
p
.
2
1
1
6
-
2
1
2
3
,
Sep
t
emb
er
2
0
1
4
.
[1
4
]
G
o
et
h
a
l
s
B.
,
“Freq
u
en
t
s
et
mi
n
i
n
g
,
”
D
a
t
a
mi
n
i
n
g
an
d
k
n
o
w
l
ed
g
e
d
i
s
co
v
ery
h
a
n
d
b
o
o
k
,
Sp
r
i
n
g
er,
Bo
s
t
o
n
,
MA
,
p
p
.
3
7
7
-
3
9
7
,
2
0
0
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
562
-
570
570
[1
5
]
H
an
J
.
,
Pei
J
.
,
Y
i
n
Y
.
,
Ma
o
R.
“
Mi
n
i
n
g
fre
q
u
e
n
t
p
a
t
t
ern
s
w
i
t
h
o
u
t
can
d
i
d
at
e
g
e
n
erat
i
o
n
:
A
fre
q
u
e
n
t
-
p
at
t
ern
t
ree
ap
p
r
o
ach
,”
D
a
t
a
m
i
n
i
n
g
a
n
d
kn
o
w
l
ed
g
e
d
i
s
c
o
ver
y
,
v
o
l
.
8
,
n
o
.
1
,
p
p
.
5
3
-
8
7
,
J
a
n
u
ar
y
2
0
0
4
.
[1
6
]
Bo
rg
e
l
t
C.
,
K
r
u
s
e
R.
,
“
In
d
u
ct
i
o
n
o
f
as
s
o
c
i
at
i
o
n
r
u
l
e
s
:
A
p
r
i
o
r
i
i
m
p
l
em
en
t
at
i
o
n
,
”
Co
m
p
s
t
a
t
,
Ph
y
s
i
ca,
H
e
i
d
e
l
b
erg
,
p
p
.
3
9
5
-
4
0
0
,
2
0
0
2
.
[1
7
]
Sav
as
ere
A
.
,
O
mi
ec
i
n
s
k
i
E
.
R.
,
N
av
at
h
e
S.
B.
,
“
A
n
effi
ci
en
t
al
g
o
r
i
t
h
m
fo
r
mi
n
i
n
g
as
s
o
c
i
at
i
o
n
ru
l
e
s
i
n
l
arg
e
d
a
t
a
b
a
s
es
,”
G
eo
rg
i
a
In
s
t
i
t
u
t
e
o
f
T
ech
n
o
l
o
g
y
,
1
9
9
5
.
[1
8
]
T
o
i
v
o
n
e
n
H
.
,
“
Samp
l
i
n
g
l
arg
e
d
at
a
b
as
e
s
fo
r
as
s
o
c
i
at
i
o
n
ru
l
e
s
,”
V
LD
B
,
v
o
l
.
9
6
,
p
p
.
1
3
4
-
1
4
5
,
Sep
t
emb
er
1
9
9
6
[1
9
]
Bak
ar
W
.
A
.
W
.
A
.
,
J
al
i
l
M.
A
.
,
Ma
n
M.
,
A
b
d
u
l
l
a
h
Z
.
,
Mo
h
d
F.
,
“
Po
s
t
d
i
ff
s
et
:
an
E
cl
a
t
-
l
i
k
e
al
g
o
r
i
t
h
m
fo
r
freq
u
en
t
i
t
em
s
et
mi
n
i
n
g
,
”
In
t
er
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
n
g
i
n
ee
r
i
n
g
&
Tech
n
o
l
o
g
y
,
v
o
l
.
7
,
n
o
.
2
.
2
8
,
p
p
.
1
9
7
-
1
9
9
,
2
0
1
8
.
[2
0
]
W
.
A
.
B.
W
an
A
b
u
Bak
ar,
Z
.
B.
A
b
d
u
l
l
a
h
,
M.
Y
.
B.
Md
Saman
,
M.
M.
B.
A
b
d
J
a
l
i
l
,
M.
B.
Man
a
n
d
T
.
H
era
w
an
,
"
V
ert
i
ca
l
A
s
s
o
c
i
at
i
o
n
Ru
l
e
Mi
n
i
n
g
:
Cas
e
s
t
u
d
y
i
m
p
l
eme
n
t
a
t
i
o
n
w
i
t
h
rel
a
t
i
o
n
a
l
D
BMS,
"
In
t
er
n
a
t
i
o
n
a
l
S
ym
p
o
s
i
u
m
o
n
Tech
n
o
l
o
g
y
M
a
n
a
g
e
m
en
t
a
n
d
E
m
er
g
i
n
g
Tec
h
n
o
l
o
g
i
es
(I
S
TM
E
T)
,
L
an
g
k
a
w
ai
Is
l
an
d
,
p
p
.
2
7
9
-
2
8
4
,
2
0
1
5
.
[2
1
]
Sl
i
man
i
T
.
,
L
azzez
A
.
,
“
E
ffi
c
i
en
t
an
a
l
y
s
i
s
o
f
p
at
t
ern
an
d
as
s
o
c
i
at
i
o
n
ru
l
e
mi
n
i
n
g
a
p
p
r
o
ach
e
s
,
”
arX
i
v
p
re
p
ri
n
t
arX
i
v
:
1
4
0
2
.
2
8
9
2
,
2
0
1
4
.
[2
2
]
Man
M.
,
Rah
i
m
M.
S.
M.
,
Z
ak
ari
a
M.
Z
.
,
Bak
ar
W
.
A
.
W
.
A
.
,
“
Sp
at
i
a
l
i
n
fo
rma
t
i
o
n
d
at
a
b
as
e
s
i
n
t
e
g
rat
i
o
n
mo
d
el
,
”
In
t
e
r
n
a
t
i
o
n
a
l
Co
n
f
er
e
n
ce
o
n
In
f
o
r
m
a
t
i
c
s
E
n
g
i
n
ee
r
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
c
i
en
ce
,
S
p
r
i
n
g
er
,
Berl
i
n
,
H
e
i
d
el
b
erg
,
p
p
.
7
7
-
90
,
N
o
v
emb
er
2
0
1
1
.
[2
3
]
Q
.
Y
o
n
g
,
"
In
t
eg
ra
t
i
n
g
Fre
q
u
e
n
t
I
t
ems
e
t
s
M
i
n
i
n
g
w
i
t
h
Rel
at
i
o
n
al
D
at
a
b
as
e,
"
2
0
0
7
8
th
In
t
er
n
a
t
i
o
n
a
l
Co
n
f
er
e
n
c
e
o
n
E
l
ec
t
r
o
n
i
c
M
ea
s
u
r
em
e
n
t
a
n
d
In
s
t
r
u
m
e
n
t
s
,
p
p
.
2
-
5
4
3
,
2
0
0
7
.
[2
4
]
Rames
h
G
.
,
Man
i
at
t
y
W
.
,
Z
ak
i
M.
J
.
,
“
In
d
ex
i
n
g
an
d
D
at
a
A
cces
s
Met
h
o
d
s
fo
r
D
at
ab
a
s
e
Mi
n
i
n
g
,
”
D
M
KD
,
J
u
n
e
2
0
0
2
.
[2
5
]
K
ü
n
g
J
.
,
J
ä
g
er
M.
,
D
an
g
T
.
K
.
,
“
IFIN
+
:
a
p
a
ral
l
el
i
n
cremen
t
al
fre
q
u
e
n
t
i
t
em
s
et
s
mi
n
i
n
g
i
n
s
h
are
d
-
me
mo
ry
en
v
i
ro
n
men
t
,
”
In
t
e
r
n
a
t
i
o
n
a
l
Co
n
f
e
r
en
ce
o
n
F
u
t
u
r
e
D
a
t
a
a
n
d
S
ecu
r
i
t
y
E
n
g
i
n
ee
r
i
n
g
,
Ch
am
,
Sp
r
i
n
g
er
p
p
.
1
2
1
-
1
3
8
,
N
o
v
emb
er
2
0
1
7
.
[2
6
]
Y
u
n
U
.
,
L
ee
G
.
,
“
In
cremen
t
al
m
i
n
i
n
g
o
f
w
ei
g
h
t
ed
m
ax
i
ma
l
freq
u
en
t
i
t
ems
e
t
s
fr
o
m
d
y
n
am
i
c
d
a
t
ab
a
s
es
,
”
E
xp
er
t
S
ys
t
em
s
wi
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
54
,
p
p
.
3
0
4
-
3
2
7
,
2
0
1
6
.
[2
7
]
L
i
Y
.
,
Z
h
an
g
Z
.
H
.
,
Ch
en
W
.
B.
,
Mi
n
F.
“
T
D
U
P
:
an
ap
p
ro
a
ch
t
o
i
n
cremen
t
al
m
i
n
i
n
g
o
f
fre
q
u
e
n
t
i
t
em
s
et
s
w
i
t
h
t
h
ree
-
w
ay
-
d
eci
s
i
o
n
p
a
t
t
er
n
u
p
d
at
i
n
g
,”
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
M
a
ch
i
n
e
Lea
r
n
i
n
g
a
n
d
Cyb
er
n
et
i
cs
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
4
4
1
-
4
5
3
,
2
0
1
7
.
[2
8
]
W
en
H
.
,
K
o
u
M.
,
H
e
H
.
,
L
i
X
.
,
T
o
u
H
.
,
Y
a
n
g
Y
.
,
“
A
Sp
ar
k
-
b
a
s
ed
I
n
cremen
t
al
A
l
g
o
r
i
t
h
m
f
o
r
Freq
u
en
t
It
ems
et
Mi
n
i
n
g
,
”
P
r
o
ceed
i
n
g
s
o
f
t
h
e
2
0
1
8
2
nd
In
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
er
e
n
ce
o
n
B
i
g
D
a
t
a
a
n
d
In
t
e
r
n
e
t
o
f
Th
i
n
g
s
,
p
p
.
5
3
-
58
,
O
ct
o
b
er
2
0
1
8
.
[2
9
]
K
ü
n
g
J
.
,
D
an
g
T
.
K
.
,
“
In
creme
n
t
a
l
fre
q
u
e
n
t
i
t
em
s
et
s
mi
n
i
n
g
w
i
t
h
IPPC
t
re
e,
”
In
t
e
r
n
a
t
i
o
n
a
l
Co
n
f
e
r
en
ce
o
n
D
a
t
a
b
a
s
e
a
n
d
E
xp
e
r
t
S
ys
t
em
s
A
p
p
l
i
ca
t
i
o
n
s
,
Ch
am
,
Sp
ri
n
g
er,
p
p
.
4
6
3
-
4
7
7
,
A
u
g
u
s
t
2
0
1
7
.
[3
0
]
H
o
l
s
h
ei
mer
M.
,
K
ers
t
en
M.
L
.
,
Man
n
i
l
a
H
.
,
T
o
i
v
o
n
e
n
H
.
,
“
A
p
ers
p
ect
i
v
e
o
n
d
a
t
ab
a
s
es
an
d
d
at
a
mi
n
i
n
g
,
”
KD
D
,
v
o
l
.
9
5
,
p
p
.
1
5
0
-
1
5
5
,
A
u
g
u
s
t
1
9
9
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.