-N } [c n -NITI LIyE #TTER

glon.T nTTd.
G'ADd' 8 OTTd'%mOTdBOOjpq t%I1YNBq(

o SSI4NlENT t

o)o ,2Ec 9CT 3oTg prAT2QDAx(: 3y3t: oocnPl p1ACI0C 6IlE(

o)o)o hmrT?2bTI9mrc

xO2? i2PYTE? NOTEBOOj ASSIGNMENT WILL OFTEN HAVE t ELEMENTSc W?ITTEN ANSWE?S: CODE ANSWE?S: AND Q2Iy
ANSWE?S% =O? W?ITTEN ANSWE?S: YO2 MAY INSE?T IMAGES OF YO2? HANDW?ITTEN WO?j IN CODE CELLS: O? W?ITE
YO2? ANSWE?S IN MA?jDOWN AND kA/Ew% =O? Q2Iy ANSWE?S: YO2? 'dCO'D%TXT LE WILL ?ECO?D YO2? ANSWE?
CHOICES IN THE Q2Iy MOD2LES FO? S2BMISSION% aOTH YO2? Q2Iy ANSWE?S AND CODE ANSWE?S WILL BE A2TOG?ADED ON
4?ADESCOPE% /HIS ASSIGNMENT DOES NOT HAVE THE Q2Iy PO?TION%

=O? ALL ELEMENTS: ,9 m9/ <9,-=x /gd bdkkr% 12T YO2? ANSWE?S 9mLY IN THE ANSWE? CELLS GIVEN: AND
:9 m9T :ELETE bELLr% -F YO2 FAIL TO FOLLOW THESE INST?2CTIONS: YO2 WILL LOSE POINTS ON YO2? S2BMISSION%

<AjE S2?E TO SHOW THE STEPS OF YO2? SOL2TION FO? EVE?Y Q2ESTION TO ?ECEIVE C?EDIT: NOT J2ST THE NAL ANSWE?%
xO2 MAY SEA?CH INFO?MATION ONLINE B2T YO2 WILL NEED TO W?ITE CODE6ND SOL2TIONS TO ANSWE? THE Q2ESTIONS
YO2?SELF% xO2 WILL S2BMIT YO2? %IPYNB LE AND ?ECO?D%TXT TO G?ADESCOPE WHEN YO2 A?E NISHED%

o)o)3 iATE 19LIbYc

) ?ED2CTION FO? THE ?ST DAY AND oz) ?ED2CTION AFTE?WA?DS FO? EVE?Y EXT?A DAY PAST D2E%

o)o)t g9W T9 hmbL2:E x92? lATH u?ITTEm mrWE?Q

xO2 CO2LD 2SE INLINE k /dw IN MA?jDOWN p?ECOMMENDED( O? 2SE MA?jDOWNS” INCL2DE IMAGE F2NCTIONALITY TO
S2BMIT YO2? W?ITTEN ?ESPONSES%

i sdw P?Eb9MMEm:E:( gE?E IS A FANTASTIC T2TO?IAL F?OM bAL/ECH ABO2T 2SING k /dw IN i2PYTE? mOTEBOOj%%
xO2 CO2LD ALSO ND VA?IO2S k /dw T2TO?IALS AND CHEAT SHEETS ONLINE%

hmbL2:E hMA4Er -F YO2 A?E STILL GETTING FAMILIA? WITH 2SING kA/Ew: HANDW?ITE THE ?ESPONSE ON PAPE? O? THE
STYL2S% /AjE A PICT2?E O? SC?EENSHOT OF YO2? ANSWE?: AND INCL2DE THAT IMAGE IN THE i2PYTE? mOTEBOOj% aE S2?E

o
TO INCL2DE THAT IMAGE IN THE bIMGr DI?ECTO?Y% kET”S SAY YO2 HAVE YO2? .o ?ESPONSE SAVED AS IMGr6+o%1NGs
THE MA?jDOWN SYNTAX TO INCL2DE THAT IMAGE IS =}+o[pIMGr6+o%1NG(%

o)3 hlPOQTANT mOTICE

xO2 M2ST CHECj BOTH S2BMISSION O2TP2T ON THE G?ADESCOPE p rrIGNMdNT t h mOTdBOOj AND rrIGNMdNT t
h <AN2Ak 4'ADING( CO??ECTLY ?EECTS YO2? WO?j AND ?ESPONSES% -F YO2 NOTICE INCONSISTENCIES BETWEEN YO2?
NOTEBOOj AND THE <AN2AL 4?ADING PO?TION: YO2 NEED TO MAjE A 1IAyyA POST: AND WE CAN HELP YO2 WITH THAT%

3 .2ESTION oc bONCEPT2Ai .2ESTIONS

rELECT THE CO??ECT OPTIONpS(% mOTE THAT THE?E MIGHT BE M2LTIPLE CO??ECT OPTIONS%

v?ITE YO2? SOL2TION AS A LIST OF ST?INGS BY ?EPLACING THE …•“

dX%c }q q[ IF YO2 THINj THE ANSWE? IS : AND }q q: q]q[ IF YO2 THINj THE ANSWE?S A?E AND b(%

3)o o)o

=O? TWO MONOTONICALLY INC?EASING F2NCTIONS FpX( AND GpX(c%

% 𝑒     IS ALWAYS MONOTONICALLY INC?EASING%

a% 𝑒  {   IS ALWAYS MONOTONICALLY INC?EASING%

b% 𝑒 3  IS ALWAYS MONOTONICALLY INC?EASING%

,% 𝑒 1  IS ALWAYS MONOTONICALLY INC?EASING%

onhlTrc t

-N }o[c Qouo 8 }q q: q,q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq ouoq(

3
3)3 o)3

vHICH ONE OF THE FOLLOWING STATEMENTS IS T?2E ABO2T SOLVING THE ?EG?ESSION P?OBLEMq

% <AT?IX INVE?SE CO2LD BE EXPENSIVE FO? COMP2TING THE CLOSED FO?M SOL2TION IN LEAST SQ2A?E ESTIMATION WHEN
THE DATASET CONSISTS OF SAMPLES IN HIGH DIMENSION SPACES%

a% /HE LEA?NING ?ATE IN THE G?ADIENT DESCENT ALGO?ITHM SHO2LD BE SET AS LA?GE AS POSSIBLE TO SPEED 2P THE
CONVE?GENCE%

b% /HE?E EXISTS A XED SMALL LEA?NING ?ATE THAT WO?jS FO? ALL THE G?ADIENT DESCENT ALGO?ITHMS%

onhlTrc 3

-N }5[c Qou3 8 }q q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq ou3q(

3)t o)t

=O? A F2NCTION 𝑒  5  . { :  ≥ 𝐺: PLEASE CHOOSE THE CO??ECT STATEMENTpS( BELOWc

% 𝑇𝑜𝑗𝑇𝑤 𝑒  5 %

a% 𝑇𝑜𝑗ℎ𝑚𝑤 𝑒  5 %

b% 𝑗ℎ𝑚𝑤 𝑒  5 %

,% 𝑗𝑇𝑤 𝑒  5 %

onhlTrc t

-N }s[c Qout 8 }q q: q,q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq outq(

t
3)5 o)5

SS2ME WE HAVE A F2NCTION FpX( WHICH IS DI0E?ENTIABLE AT EVE?Y  ≥ 𝐺% /HE?E A?E TH?EE P?OPE?TIES THAT
DESC?IBE THE F2NCTION 𝑒 c

po( 𝑒  IS A CONVEX F2NCTION%

𝑏𝑑 𝑤} 
p3( vHEN  5 2 : 𝑏𝑤 5 .%

pt( 𝑒 2  IS A GLOBAL MINIM2M OF 𝑒 %

vHICH ONE OF THE FOLLOWING STATEMENTS IS W?9m4q

gImTc xO2 CAN 2SE A FAIL2?E CASE TO DISP?OVE A STATEMENT%

% 4IVEN po( AND p3(: WE CAN P?OVE THAT pt( HOLDS%

a% 4IVEN p3( AND pt(: WE CAN P?OVE THAT po( HOLDS%

b% 4IVEN po( AND pt(: WE CAN P?OVE THAT p3( HOLDS%

onhlTrc t

-N }oz[c Qou5 8 }q*q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq ou5q(

3) o)

-S THE FOLLOWING STATEMENT T?2E O? FALSEq 4IVEN ANY CLASSIE? WITH jNOWN PA?AMETE?S 𝑡 AND : ITS DECISION
BO2NDA?Y IS ALWAYS F2LLY DECIDED%

% /?2E

a% =ALSE

onhlTrc 3

-N }ot[c Qouf 8 }q*q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq oufq(

5
3)7 o)7

-S THE FOLLOWING STATEMENT T?2E O? FALSEq 4IVEN A LINEA? CLASSIE?  5 SIGN 𝑡  : IF THE DECISION BO2NDA?Y
IS GIVEN: THEN ITS PA?AMETE?S 𝑡 AND  CAN BE F2LLY DECIDED%

% /?2E

a% =ALSE

onhlTrc 3

-N }o7[c Qou7 8 }q*q[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq ou7q(

3)s .2ESTION 3 bONuEv

-DENTIFY THE CONVEXITY FO? THE FOLLOWING SIX F2NCTIONS pAhF(

v?ITE YO2? SOL2TION AS A LIST OF ST?INGS BY ?EPLACING THE %%%

dX%c Q3uCON_dX 8 }qAq: qBq: qCq[s Q3uNONuCON_dX 8 }qDq: qdq: qFq[ IF YO2 THINj F2NCTIONS pA(: pB(:
pC( A?E CONVEX AND pD(: pE(: pF( A?E NONhCONVEX%

onhlTrc 5

-N }on[c Q3uCON_dX 8 }qAq: qCq[ n r#k2/-#m

-N }3z[c Q3uNONuCON_dX 8 }qBq: qDq: qdq: qFq[ n r#k2/-#m

-N } [c G'ADd'%C>dCjpq 3q(

3)e .2ESTION t Q4lIN AND Q4lAv

N 2NjNOWN ESTIMATO? IS GIVEN AN ESTIMATION P?OBLEM TO ND THE MINIMIyE? AND MAXIMIyE? OF THE OBJECTIVE
F2NCTION 0 𝑡 ≥ .+ [c


 𝑡𝑦 + 𝑡𝑎  5 A?GMIN𝑛 0 𝑡+ A?GMAX𝑛 0 𝑡

/HE SOL2TION TO dQ% o BY THE ESTIMATO? IS 𝑡𝑦 + 𝑡𝑎  5 .+ .%

4IVEN THIS INFO?MATION: PLEASE OBTAIN THE VAL2E OF 𝑡− S2CH THATc

𝑡− 5 A?GMIN𝑛 =. {  − LN 0 𝑡[

YO2 MAY INSE?T IMAGES OF YO2? HANDW?ITTEN WO?j IN CODE CELLS: O? W?ITE YO2? ANSWE?S IN MA?jDOWN AND
kA/Ew%

onhlTrc 5

sxP, xnt. lru,. g,.,: .,Pi ahle Tghr T,wTp

𝑡− 5 A?GMIN𝑛 =. {  − LN 0 𝑡[

5 A?GMIN𝑛 ={ − LN 0 𝑡[
5 A?GMIN𝑛 ={0 𝑡[
5 A?GMAX𝑛 =0 𝑡[
5 𝑡𝑎
5 .

3)n .2ESTION 5 ,ECISION rT2lP

-N THIS P?OBLEM: WE WILL PE?FO?M A BINA?Y CLASSICATION TASj ON A MODIED -?IS DATASET% /HIS MODIED -?IS
DATASET HAS 3 CLASSES AND IT IS SPLIT INTO A T?AINING SET 𝑅snAgigie WITH ozz DATA POINTS AND A TEST SET 𝑅sars WITH
z DATA POINTS% dACH DATA POINT X+  HAS A FEAT2?E VECTO? X ≥ ∑ AND ITS CO??ESPONDING LABEL  ≥ ].+ |%

p-N FACT: THE O?IGINAL -?IS DATASET HAS t CLASSESc z FO? rETOSA: o FO? uE?SICOLO? AND 3 FO? uI?GINICA% gE?E FO?
BINA?Y CLASSICATION TASj: WE COMBINE rETOSA AND uE?SICOLO? TOGETHE? AS  5 { AND LABEL uI?GINCA AS  5 (

gE?E WE 2TILIyE A DECISION ST2MP TO SOLVE THE ABOVE BINA?Y CLASSICATION TASj% /HE DECISION ST2MP WO?jS AS
FOLLOWS pFO? SIMPLICITY: WE ?EST?ICT O2? ATTENTION TO 2NIhDI?ECTIONAL DECISION ST2MPS(c

f 4IVEN THE FEAT2?E VECTO? X: THE FEAT2?E INDEX : AND A TH?ESHOLD 𝑆 𝑔: THE CLASSICATION F2NCTION IS DENED
BY  5 𝑒 X+ + 𝑆 𝑔 ASc
 ℎ𝑒 X  ≠ 𝑆 𝑔
𝑒 X+ + 𝑆 𝑔 5 )
{ 𝑠𝑔𝑜𝑡ℎ𝑟,
WHE?E X  ?EFE?S TO THE hTH FEAT2?E IN X%

7
 f /HE E??O?  ON DATASET 𝑅 5 ] X𝑓 + 𝑓 | IS DENED ASc

 𝑖
5 ∈ o𝑥𝑓 ∗ 𝑒 X𝑓 + + 𝑆 𝑔(
𝑚 𝑓54

WHE?E 𝑚 5 𝑅 IS THE SIyE OF THE DATASET 𝑅% /H2S: WE CAN OBTAIN T?AINING E??O? snAgigie ON T?AINING SET
𝑅snAgigie : AND TEST E??O? sars ON TEST SET 𝑅sars %

aASED ON THE DECISION ST2MP ABOVE: WE WISH TO 2SE AN ALGO?ITHM TO ND THE aErT eEAT2?E Im:EX − AND
aErT TH?ErH9L: 𝑆 𝑔− ON T?AINING SET TO C?EATE A …BEST“ DECISION ST2MP: IN A SENSE THAT S2CH DECISION ST2MP
CAN ACHIEVE THE L9WErT T?AImIm4 E??9? −snAgigie %

LGO?ITHMc h -NITIALIyE −snAgigie 5  h FO?  8 z: o: 3: tc h FO? TH?ESHOLD 𝑆 𝑔 8 MINpX (: •: MAXpX (c h
bALC2LATE snAgigie h -F snAgigie l −snAgigie : THENc h − 5  h 𝑆 𝑔− 5 𝑆 𝑔 h −snAgigie 5 snAgigie

f 92TP2T BEST FEAT2?E INDEX − : BEST TH?ESHOLD 𝑆 𝑔− : LOWEST T?AINING E??O? −snAgigie AND CALC2LATE ITS
CO??ESPONDING TEST E??O? −sars %

3)n)o .2ErTI9m f)o

1LEASE COMPLETE THE FOLLOWING CODE pYO2 SHO2LD ONLY INSE?T YO2? CODE IN THE %%% PA?T( TO IMPLEMENT THE
DECISION ST2MP ALGO?ITHM ABOVE%

onhlTrc o7

-N }35[c )}nmFgf -NkINd*ACjdND%FIG2'duFO'MAT 8 "'dTINA"

glon.T m2lou r mo
glon.T l ToknTkgapouoknT r okT
F.nl rikd .m glon.T DATArdTr

-N }3f[c n -RIS D T SET)

I'Ir 8 DATArdTr%kOADuI'Irp( n kO D -RIS D T SET)

w 8 I'Ir%DATA n /HE SH 1E OF w IS pofz: 5(: WHICH ME NS

n THERE RE ofz D T 1OINTS: E CH D T 1OINT
n H S 5 FE TURES)

n >ERE FOR CONVENIENCE: WE DIVIDE THE t jINDS OF FLOWERS INTO 3 GROU1Sc

n x 8 ho c rETOS pORIGIN L V LUE z( s _ERSICOLOR pORIGIN L V LUE o(
n x 8 o c _IRGINIC pORIGIN L V LUE 3(

n /HUS WE USE pIRIS)T RGET g o)f( TO DIVIDE THE T RGETS INTO 3 GROU1S)
n /HIS LINE OF CODE WILL SSIGNc
n xuORIbI[ 8 /RUE pWHICH IS E.UIV LENT TO o( IF IRIS)T RGETbj[ g o)f p_IRGINIC (

;
 n xuORIbI[ 8 ! LSE pWHICH IS E.UIV LENT TO ho( IF IRIS)T RGETbj[ l8 o)f prETOS s _ERSICOLOR

xuO'I 8 pI'Ir%TA'GdT g o%f(%'dr>A1dpho:o( n /HE SH 1E OF x IS pofz: o(: WHICH ME NS

n THERE RE ofz D T 1OINTS: E CH D T 1OINT
n H S o T RGET V LUE)

n vE THEN CONVERT /RUE ND ! LSE IN xuORI

n TO o ND ho IN xc
x 8 N1%yd'Orppofz: o((
x}xuO'I 88 s.2d[ 8 o
x}xuO'I 88 e krd[ 8 ho

wuANDux 8 N1%>rTACjppw: x(( n rT Cj THEM TOGETHER FOR SHUFFLING)

N1%'ANDOM%rddDpo( n rET THE R NDOM SEED)
N1%'ANDOM%r>2FFkdpwuANDux( n rHUFFLE THE D T 1OINTS IN wu NDux RR Y

1'INTpw%r>A1d(
1'INTpx%r>A1d(
1'INTpwuANDux}z[( n /HE RESULT SHOULD BE LW YSc b f)e 5) o)3 z)3 ho) [

pofz: 5(
pofz: o(
} f%e 5% o%3 z%3 ho% [

-N }37[c n ,IVIDE THE D T 1OINTS INTO TR INING SET ND TEST SET)

wur>2FFkdD 8 wuANDux}c:c5[
xur>2FFkdD 8 wuANDux}c:5[

wuT'AIN 8 wur>2FFkdD}cozz[ n rH 1Ec pozz:5(

xuT'AIN 8 xur>2FFkdD}cozz[ n rH 1Ec pozz:(
wuTdrT 8 wur>2FFkdD}ozzc[ n rH 1Ec pfz:5(
xuTdrT 8 xur>2FFkdD}ozzc[ n rH 1Ec pfz:(
1'INTpwuT'AIN%r>A1d(
1'INTpxuT'AIN%r>A1d(
1'INTpwuTdrT%r>A1d(
1'INTpxuTdrT%r>A1d(

pozz: 5(
pozz:(
pfz: 5(
pfz:(

-N }3s[c :dF i2DGdpA: B(c

qqq
iUDGE FUNCTIONc op =8 B()
"ETURN o IF =8 B: OTHERWISE RETURN z)
qqq

e
 n ad4-m r#k2/-#m
gF A =8 Bc
.dT2.m o
dkrdc
.dT2.m z
n dm, r#k2/-#m

:dF FuDdCIrIONurT2M1pX: i: />(c

qqq
,ECISION STUM1 CL SSIFIERc FpX: J: /H(
/HIS FUNCTION SHOULD RETURN ho OR o)
qqq
n ad4-m r#k2/-#m
gF X}i[ g8 />c
.dT2.m o
dkrdc
.dT2.m ho
n dm, r#k2/-#m

n ] LCUL TE ERROR GIVEN FE TURE VECTORS w ND L BELS x)

:dF CAkCud''O'pw: x: i: />(c
d 8 z n r#k2/-#m
N 8 kdNpw( n r#k2/-#m

Fn. pXI: YI( gm yI1pw: x(c

n >INTc 2SE JUDGEp( ND FuDECISIONuSTUM1p(
n ad4-m r#k2/-#m
d .8 i2DGdpYI: FuDdCIrIONurT2M1pXI: i: />((
n dm, r#k2/-#m

n >INTc REMEMBER WE W NT THE VER GE ERROR)

d 8 o%z a d 6 N n r#k2/-#m
.dT2.m d

n < IN LGORITHM)
O1TuduT'AINING 8 o%z
O1Tui 8 ho
O1Tu/> 8 ho
Fn. i gm }z:o:3:t[c
Fn. /> gm N1%A'ANGdpwuT'AIN}c: i[%MINp(: wuT'AIN}c: i[%MAXp(: z%zf(c
duT'AINING 8 CAkCud''O'pwuT'AIN: xuT'AIN: i: />(
gF duT'AINING l O1TuduT'AININGc
n >INTc U1D TE BEST J: /H: ND LOWEST TR INING ERROR HERE)
n ad4-m r#k2/-#m
O1Tui 8 i
O1Tu/> 8 />
O1TuduT'AINING 8 duT'AINING
n dm, r#k2/-#m

1'INTp"91TIMAkc ia8b{: />a8bcp3F{: duT'AININGa8bcptF{: duTdrTa8bcptF{"%FO'MATp

O1Tui: O1Tu/>: O1TuduT'AINING:
CAkCud''O'pwuTdrT: xuTdrT: O1Tui: O1Tu/>(((

n
91TIMAkc ia8t: />a8o%sz: duT'AININGa8z%zoz: duTdrTa8z%ozz

-N } [c G'ADd'%C>dCjpq 5uCODdq(

3)n)3 .2ErTI9m f)3

rHOW THE HISTOG?AMS OF EACH FEAT2?E AND D?AW A LINE FO? BEST DECISION ST2MP% aESIDE THE LINE PLEASE SHOW A
TEXT OF LOWEST T?AINING E??O?% xO2 MAY 2SE "duT'AININGac \{" AS THE ST?ING TEMPLATE%

onhlTrc 5

-N }te[c 1kT%FIG2'dpFIGrIyd8po3:e((
Fn. i gm 'ANGdp5(c
wiuT'AIN 8 wuT'AIN}c:i[
wiuv>dNuxzuT'AIN 8 }wiuT'AIN}I[ Fn. I gm 'ANGdpkdNpwiuT'AIN(( gF xuT'AIN}I[ 88 ho[
wiuv>dNuxouT'AIN 8 }wiuT'AIN}I[ Fn. I gm 'ANGdpkdNpwiuT'AIN(( gF xuT'AIN}I[ 88 o[

1kT%r2B1kOTp3: 3: i.o(
1kT%>IrTpwiuv>dNuxzuT'AIN: kABdk8"!dAT2'd b{: x8ho"%FO'MATpi((
1kT%>IrTpwiuv>dNuxouT'AIN: kABdk8"!dAT2'd b{: x8o"%FO'MATpi((
1kT%XkABdkp"!dAT2'd b{c b{"%FO'MATpi: I'Ir%FdAT2'duNAMdr}i[((
1kT%YkABdkp"!'dQ2dNCY"(
1kT%kdGdNDp(

gF i 88 O1Tuic
n >INTc USE 1LT)1LOTp( ND 1LT)TEXTp( TO 1LOT THE VERTIC L LINE ND TEXT)
n ad4-m r#k2/-#m
1kT%1kOTp}O1Tu/>: O1Tu/>[: }z: oz[(
1kT%TdXTpO1Tu/>: oz: "duT'AININGac b{"%FO'MATpO1TuduT'AINING((
n dm, r#k2/-#m

1kT%r>Ovp(

oz
t dND O8 t

t)o r2alISSION

<AjE S2?E YO2 HAVE ?2N ALL CELLS IN YO2? NOTEBOOj IN O?DE? BEFO?E ?2NNING THE CELL BELOW: SO THAT ALL
IMAGES6G?APHS APPEA? IN THE O2TP2T% /HE CELL BELOW WILL GENE?ATE A yIP LE FO? YO2 TO S2BMIT%

1LEASE MAjE S2?E TO SEE THE O2TP2T OF THE G?ADESCOPE A2TOG?ADE?% xO2 A?E ?ESPONSIBLE FO? WAITING AND
ENS2?ING THAT THE A2TOG?ADE? IS EXEC2TING NO?MALLY FO? YO2? S2BMISSION% 1LEASE C?EATE A 1IAyyA POST IF YO2
SEE E??O?S IN A2TOG?ADE? EXEC2TION%

-N } [c G'ADd'%dX1O'Tp1DF8e krd: FO'CdurA_d8s.2d: '2NuTdrTr8s.2d: FIkdr8}"IMGr": "'dQ2I'dMdNTr%TXT"[(

oo