multivariate methods 3

- X X

-
-

1
X

i
x X

DDD
AAA
o
taken from each point
multiple measurements are
second feature ,
=>
D ->
xixI
[Xil xiz
feature
↳ pth
- -

tIR
-

* =
=

Xi
ithdate ↳ first feature
velve
= terget
point .

S
label y
=

y=
class
N ↳
EIR
*

y:
51 2
,
---
-
k3
yi)3=
,
*
x =
2( * : ,
=
1
gitRt
*

"in I of t
X12 --- XIDA X 1 T
X21 X 22 --- X2D- * 2
*
-
=

i
XNz --- XNx
-
* N Nx1
N XD
 " **** ****
X

x x X

x & X
x
X I I
X I

[-
& I

] ro " 2) r
12 = 81z=5
-O

univeriate => x WN(x ; y, (2)

--
↑
[xi IX I
T = I
muliorate
*
N
=> * ~N(
*;,
-
5
- coveriace
matrix

xin
xx
~2
We
⑱ i
-
DX1
1 =

N
meovector
T
IxI
N
M -

Y)x) * ) , xx

.
⑤ =

DXD M
 N(x ; y ,
c) = ex -

-]
)!x]
p 8) exp[ ( -

N( :
- -

(5)
=

, -
* xD Dx 1
en IXD

density

-
-
determinant |x I
/x I

wher D=

[02]
-
-
I
exp[-]
S =

,x
n
(2πst-
2

-
1
1. exp
[ **]
-

-
2482
 class conditional
Multivariate Parametric Classification density
c) ~N(x ; ( ,)
X2

p(x1y
-
↓
=

i
: I
, S
1
exp[- (x -
x = xx -

+z)
:
-

: 5)
O BX
a

leg[prly 3)
.

08
XI
↑ )]
log(py
+
=

TXleg)(c)
-

9(
=

X Bx x Ba-a
=

e) =
+

)]
-
. .
-

y + leg[(y
- (x- -
2
=

9((X) 4-leg(2π) -
-
↑
= -

W
-

⑧
en -1 - ~ 1
men
-
1
x= ·Pr- Er
-

x
w

bet of
~ x +

C
-

parometers I
·
- -
~

[
-

atXb btXa
[41)]
+ k - 1 =

k t
k D -
if X is squere
+ .

=
.

e ne p-Bilities A symmetric
coverience proble
bT AT
mean
rectors metrices (b)
to
/
=
.
↳ba
 order
M
1(y= c)] =
=> first .

B
=
.
"1(yi 1
=

c)
i
order
=

( = Y()(xi 4) T 1(yi c) x
second
polynomial
5.
=
- -
.

1(yi c)
=
=

constant
x
(y= c) frequency of
Pr(y N class
c)
= .

=
=
=
=

N A
Wa ?
0
=

0T

⑦J
wcx Nco
X1 x
+

90x)
W
C ?
+

C
=
x .

Nc
=

S
.

-
Nco= ?
2 1
I
T

2 legy x)
-

Ec
↑c
log((c))
.

+
log(2) -

!Y
-

↓Y
 91(x) xY We w ,x w10

3
x
+
+
=
. .

92(X) = xT IN2 .
.
x +

W2 .
x +
Y20 pick the maximum
i
i ! one
i
xt Y1x x, x Nk0
31(x)
+ +
.
.
=

--
when K =
2
T
wi
xTN1
we x
+

x
91(x)
+
.
= .

xT /2 x w x +

wz o

92(x)
+
.

= .

=
) =

x +(W1 (2)
ne
-
.
x +

(x1-w2) x
-
.
+
(w10 - w20)
e

W Y No

30 Elstclass
+ wx
-

q(x) =
xYW . X + w .

↳ < and class