Record

Motivating example why prior important apart from data

• •

Bayesian approach
•

Beta binomial Normal

gamma
- -
.

• Parameter :
f ( µ , -0 ) → 0=402

,%Y✗¥¥yaÉ%÷%
§
" Brion : *
( p) f- Beta (a) b) it
/v10) → N / µ .

' H'
gammato.X.si?o
" ' °
e'"
_

*
(a) →

Likelihood of 44 / a) a 0% e- 414¥10
observing
2.
data µ,
' " "
L( ✗ Ip) a
p G- f) a-
surmise

3. Posterior for ( µ . ,

a- ( PII ) Beta Cats , b-in -

s
) IT ( MII / 0) - N (Mif)
 E- ÷+
.
no

M=÷
¥ =
÷ +
¥

IT ( OIE ) -

Gamma ( M A) ,

8, =

% -180 2

"
1--1 + E
ñ
FEET →
÷ ""
4. Posterior
inference .
Simulate Q
,
µ/ 0

Mean =

a¥b ,

van =

? (a -14+7 ,

Credible
intervals qbeta

5. Prediction
for X
P / ✗ new ≥ t ) = F-
[ P( ✗ ≥ᵗ / ii. 0 ) ]
new
P( 2) E[Pl✗new=2 / ☒
✗ = =
new

Simulate 0
from gamma /HAD
simulate
f from ( Pl ✗
old )
simulate ii. given 0 from N( µ 2,4
simulate ✗ Binomlh f) , ,
-

new
NIM %)
.

simulate
"
✗ new ,
Normal model : Unknown 0 known
µ
.

Predictive distribution

✗
new / MUN / u ,
04
↳ known

prior
or Posterior
Ms N( M ,Z ]
, ,

then

✗
- Normal (µ , ,
I ,ᵗ+)
new

F- [ ✗ new
] -
F- [ F- [ ✗ new / u]]

r[ ✗
new ] = F-
[ v( ✗ new In )]+v[ F-
( ✗ new / MD
7¥ ÷
group group .
 µ Known ,
T unknown .

matrixops.CI/)---x- ( symmetrical]
✗

LimauRegression_
Prediction
Bl R2 Bk
.

[¥;)
. . .
_ .

Data
y
H
, Nz ' " ' "

NK xp when f-
I
N' 2
2 : ← ✗
nxk
3 ?
÷ .

↓ M2
~

Ynxl
 For owe case , Xm , =
I Am C- [ 1) ] n

Yi
=
Bit Panis +
Baig + . . . +
Éᵈ N / 0,09

"
'

Least Squares
"

Minimise É [Yi
,
- +
B. % -1
. . . +
Brain)]
%
↳
Y XB
✓ error sum
of squares

@
-

VTV = Min -
✗ -

xp)
B
(sum of
squares)
?⃝
Fact .

( Proof is exercise )

④ -
✗ B)
"

( Y -

✗ B) =
-
MY Dfs M) - + YT ( I - ✗ 1×-1×5 F) Y
'

[idea

{
a
match coeffecient]
# ☒ TY §
→
Mi ✗ =

PTDP
D= ✗ TX
2 MTDB

analogous to E@ i -
MY =

(M -
# Minimum is at B- -

'

Fact § : ✗ 5 #
Ya , ,
"

v11 ) =
T2 ( ix) kxk
 Linear
Bayesian Regression .

§ⁿ( y ;q#ÉÑn)2
Data likelihood .

n
-

11111,02 ) ✗

¥g ) e
'

202

t.lk/f,o)aoMze-H-xB5(Y-
✗ B) univariate
•
2

Adt
Prior ( p ,O)
: -4%515%-8) ¥géᵗEˢ!
¥!
2

plonnnlq.is ) &ñ¥s→
mutivariattdf e-
0 Yamamoto A) 0%-1
" 0

÷,%=É
e- ↳
is ◦

, a

*
p,
B2
 '

HE ftp.j
-

Posterior for (B) Of ( B-

0% (Y-✗BIDO e- 0%-1 ✗ 0

¥
2
_
◦
-

/ 1,0 / Y )
,
✗ ✗ ×
IT ,

LMAO ) it
/p ,
0
)
" ( o)

ñ( 01¥ B) in

gamma (81/4)
E' G- PD
,
'
← B)
e- ? ?
( B / 1,0)
A-
✗ =
E
Normal,
=

+ is
2 ,

"

¥151s ] [¥¥-&+é's]
"

q=[
[ ¥-2 ]
' '

si = + É __
sum
of precision
For univariate
,

N (M T2 )
Ms . _ _

,
Rn u
,

101M$
}
simulate On IT / Old ) exact approach to simulate

simulate Mutt 1m10 ,

For multivariate IT / MOI E.)

,

!!¥ } aftoo " mate

{
simulate -
approach to simulation .

simian

[ aeibbs
Sampling g.
g. 04×-1×51
=

Zellner's Poison
↳ like Yr
g