Scribd
Upload
25 views

A Depth Intuition of Linear Regression Algorithm - )

The document discusses minimizing a cost function by adjusting parameters through an iterative process. It explains that the cost function value can be decreased by: 1. Taking the negative of the slope of the cost function at a point. 2. Updating the parameter value in that direction. 3. Repeating until the slope is near zero, indicating a local or global minimum has been reached.

Uploaded by

chinmaysinghsocials
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
25 views

A Depth Intuition of Linear Regression Algorithm - )

The document discusses minimizing a cost function by adjusting parameters through an iterative process. It explains that the cost function value can be decreased by: 1. Taking the negative of the slope of the cost function at a point. 2. Updating the parameter value in that direction. 3. Repeating until the slope is near zero, indicating a local or global minimum has been reached.

Uploaded by

chinmaysinghsocials
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Linean Reyemn

Aipevdant fatus ad
Gontineons. Namly ( , 205, 1o2, 103, 40,
Datant 6so, - - - ) .
Salay
Datnet

Bet ft line

he main im e best tt Achual data pernts


oUos betwe
te
minimzy
predicd whach is to adiustbatueeh
and that tte
ine Con ditne mlaimum
t Audichd dat poink,
Achuad
Lot ft line is
b ondentond what a
Need t eqvatin )
,slope
ha) = + 6(x).
LSlope or ton e
intept

Unit moe mert n -ayis what i


te meve went uy-ay

when we
best t ne Chonges (#
best tted (ine

(ems iS huge.)

Do ngod to mnim te eoS1, how do we

do tthat,???7
Rrictd. Acual
Cost fonchm

need to

Name Casthuneter

(
Ahuad depeudont vonuable pont .

takoept sl

MSE
2
3

New,
Aaummine ightlly 0, },
3

when,
3 X=2
he 2) = 2
3
ie' Compite ost hunttia ,
J (e., 6.)
2

-(2-2)-(3-3)|
-

No e,)

3t

(Go)

o-5

e)

2=1

=2
Nw i i clonge my caludoe lo

o-)+ (o-+to-)
3
J ) 46

Je,) = J

No Capaa the -ue


|,-0,
Pan |20
2 2

3 3
22 4

A S

Gtradient7 pla )
3:5 Dogtant
inhirle hen,(e,
2
(D-5,
1-1b) ,- , J(e,) =o
19

o,=0,J(e, )=466

Dike this plt te


Vallw t ,)Jo) i4 kpm a porabila

Same
So in oley
ngsd tto chone , +o, vale.
wwch ds tee
A mochonsm hod be tolltwed
.??’ Gnergpnla tgam Coptimiy the e, vale
Gryodsont

ttis alpld bs
why uprd
paoaboto.35
tha cast
e
Pepot uni1 GMgeno

thd eut te sjepe


at tte phular
)

d =y leaAne rate

slibal mhima 99
the
reoch
tele s that,
Convergenle eq wation
The ubtrecttd uith
kunld beke the
leanierae (a) and
praduet t! t teie point
the tunthion at
deivative e Cot
Can be o, ,b,)
J(o)

slepe)

L qlobal
minima

How to ind it (-v) slope oY (tve lope

Draw tangont to te oint

te tangent fa Ces do on wands,


do o
Riqht hond side
a ve)slope walue
Na the

- <-ve valu
: 6;
Thiu illineneas ;
hond aide
Right
it Lctomes a (+ve) aleze vale.
tt Connengern le epu
Now
f+ ve value
b,O; -
[value]Y ihis wil datreae
Jte)?

glotel
(decnsase
)
Rate (a docldes
Jentng
larae
wee veS
les and Ieo modal reothed glatal mihima
globl mnima doon.
Amall
Vale

model leans all


latge o the the
glstal minimo miai Ma

You might also like