Reresim 1

ML- algetm redeted Une

Superisel ML Aetuat
Doaset Calalate fw.

b 7-mt
(-) -+tanget
Err(E) = e o r /Loss ’ minimle.
(-mni-b)..
Chonge in E fr
Chbmgt in b.

(G-mxi-b)'o<
Liuor
’(-m,-b)"=0
Dala
> ( - mx;-b). (-)=0
> y i - m -b o (+)
1020 b
-mi-b

be2j n= 4
2t2+242
b+bt.b
Y+Yt..-Yn
Dote -0 Pmely LinPrw Teal bfe x
Yeal bfe -v ’ y- m-be0
’ b= y-m7.
Best fit lineT Cest
depondntng)
(Pa)
Asume hafind
e m+b
M> slope ($-M-6)=0
be y- inteopt’ (wedtg).
(ofset
3
y* independent yaiable (up) (o
dapendent variable 2( - mNi-9 +m)(-+)=o
(p). (-)-m(ai-+Je(i-)
indepen dentE
’
(*;-)")= o
(6-P(;-4) - m
enpnne
m= (-9) (i-)

Pg m*enpeñena Jireet fomula

plg mo onpeHenezo/teskes b> 9-mt
m (y )(

feshos phgzo... t pssi ble

So b( offse eguied.
tmget’ fnd out onea prsHn n &b: Ihrodien egrasen he
direiien f stapest asun
finding
m&b
wos’
ensure w one at eeal
peixl
Cosed Jom seahon [Non- Clesat forn setaHio
Direl- fTmula Apperinnatm Bascd goromtead to be the
OLS technique (ordiney Last Spune) Gndtent Desunl minimoa.
iromnple Los Surfuu ’
= -bt20Vy-yo
Exat solulian
Sk lean ’ Liny Regresim Cm intemaly uue
SGD Resim Cam- OLS 4elhrigu.
 lavealm
k-- Multpu La calulan,
Kresim Matias, Eor tn()’ "Smple
Meom Abseute EoY
E--$Y
sds Not dhible a o
ady: Outiey ftect not se muh
MAE unit = p cof unit
Mem SquL Fm
ATo8
(7-) T-8) (y-$) ---.. ao=
edu Difforentable at o.
outlier ofet
MSE wmit # op uumit
Reot Meom Squu F
VMSÉ
ae: Difenenhablu ot o
RMSE nit p mi
=(A 8)--- (aB)= 6TAT]
dss outber otteet -hy C
---- te= ATS]
R Sure
-meon

$Se
SSm
2(.m)( ,1)
’ best
’ bad w
predicE = J8-2aw+u)
nwost
Mem

+ ( w ) =0 (mabix).
’ 0- 2
i Lezenig
i =m+b

(w) (y («*a)")

i/p Seotne =m
XT= (mn)
(mt, n)(n,me)
(mal, m)
(e)-n, mi)
W

W3
(, ma >(met 1)
-(m+)
(a,? ( m) (n1,1).
O (oxn x(n+)) o()
e

Suppase e feature 999 total daa 1000

2 mputatim requinel l000 X1000 ^|000
’ 1000 o00000
 Multpl LR
Gvodent Decunt ... wty?
maix multipleahon nsn ’ timu Cmpliy oa).
1 0rder optimvzohn clgoitbm
dApprah for y m+b)
D Select Tamdon bk m tal)
DAnd slope f b& m
D Update b R m m= m- nslope -±[(o-3)+ (6-)]
b b-. slope
When to stop
t0D0 time epochs.

|L (-) ±[-2(y-&)-2(4)T
=t[-z ()+(h-9)I
*2(i-mt; -b) (-) 2(-mni -b)(4) = # )+(3)]
=-2 (-m-b) e-2 2 (-m-)(0 -[(-)+ (%-) +(0-)+... (th-gn)]
/meom

clas CaDRegyzser:
det -init.. (selt, tearnin note, epohs): (-Bo-Aitn -bsn) -x)]
selk m=
self. b = = 2(-fo-B n B, Xn)(-i) t 2
self. lr= teorning.rot -t[-2(8-9,) (*) + 2(4-)()]
self. epochs epochs =t[-2 ((-9)(*1) + (h-&)(n)
for i in romge( selt. epochi):
tos-slope-b-2*np,Sum (g- selt m* Toel ()- self. b)
oss-slope-m -2*np-sum-sef.na XTovel()-celp rove) -5)(1) .
slt.be selt. b (ses tr* lors-slepe-)
Sut.m= suf. m-l slt.tr los-sope -m)
prinb( setm, self.b)
det prediet ( selk x):
| retum elf. m*sels b
a - -%2(-)(Win).
g4.itGoRgrz4or
(Xrom(o-0o,
-tran)so)
Shape
241 2X1
Sualer
3 4

231

form

Crrodlent Derunt - (-)()

(Spehaoth]
updete b1 update
n updte psih
thye Dat
tut cmmgene , - [9-8) .bain.
'boteh sloo Hug Dat
Shhosie fast
othma'se fot
 Polynomiat RereuiY*onmtb stoeterug
L (*-94)"+ Am
Nulttpla LR
m Consttt bla
Cxpr porammil7
dgreer to d 3

T35 |oD|

deyru>2 |°xxx3 Y 20 be Censtt

Om
|o351215 o3s 2s 285|

2Àm = o
2 (i-mni-g+m)(-N+1)+
2- m-g4m) (-)+ 2m= 0
degreu= (*-3)- m (u-)
(ai-)] =o (bhside mlt pt )
X Am - 2
Am-*-)(i-*) - m
(1-7)"]=0
Am(-3) (*%-5) t m (*-)o
Amt m (u-) = (%5) (-)
Bias Vom)nee Trade -oft Li’ Simple LR
A=0 -.

1Bias mchngeb alio ehany

OHigh biau: Unable to tly copure the mlatnship in
hroin data
AD OLow bias : H the obility to tly Coyture thu yelahimhi w w+wt... w
in tran data
Vanionca (og + w)
I hgh Vorionle: Troinse yo
’Tutse Vor= Hihoz
[h-) ) -..
e n r 60 eTe +A wT
=(9-8)-) +AAw w
troin data +
otet dota -(8-$). (y-9)
Rejec) Bias Bias
Voriomu Variome4 Noniomue t R i u

eularizahm
A cmmque in ustch e odd smu enta tnfo
Se fhawe am eduu oreftng
T= Hmtity mahin
. Ridga Pyulmizohon 41) Seale value
ba
2. Lao
Rgulontotm ()
3.Elosie Net Regulitohton t )
Y= mn+b
oton dah
L3(-)* Aw
(8-8. (9-8) +
(-(w-1)) +ATw
* (-(w-)
*(-1( - ) (w) tAww
(uu)-y(rw-)+AwTw
=o+(s)" =(wNT-y)(w-4hwTw
-£(waw-
( - )(w)(u)
wywtyy)+ia wTw
4 [wxTx w- 2wy+Y]+wTw
Low en
 FAS
m’)

L ($-m;-b)°+A m
L- C-)+A w

Shrtnk vae
W . Wn
w, W Wn 0
l(-5) (-1)- m(*-)°]+Ao

A=0 (ri-)"= (-)(-)-»

re feted more
(8-3) (*-*)-à
:(oet of )
A=o 516. 69 S242
owJump
Liw Troion
Bias voniome TTrolestt
Oveft Bias
L= (-ns-)-Ain odd 'u fr dltette help
( - m- 3+ m)- 2·m zo
Nee
Voiomu
=

L(-)+A w
m2
(9-5)(-) +A
(4-)

Pigde
L=(*-%)+Au MSE
oftuteh

OLs /Dlret formula sot

yreion cof. ofteuted bias

Bios Nojomee
mpo m

3 we
b-5- mz

mO
|La (8-9)* + alo+bo
4.tto
mo
(h-3)(-*)-> ()) mo
(9-)

neimesat O Malidonai pruent thun we Elnhenet

Li belngs
dunomnah.