cs 189 final review
DESCRIPTION
CS 189 Final Review PPT for class CS 189 in BerkeleyTRANSCRIPT
CS#189#Final#Review#Session#–#Part#1#
University#of#California,#Berkeley#
CS#189/289#IntroducBon#to#Machine#Learning#
Advice#
1. Make#a#table#
• Rows:#ML#techniques#(SVM,#logisBc#regression,#etc.)#
• Columns:#
• Sources#of#overfiQng#
• Parameters#
• HyperTparameters#
• Algorithm#convergence#
• Common#opBons…regularizaBon,#kernelizaBon,#
etc.#
Advice#
2. Be#very#comfortable#taking#a#gradient#of#linear#algebra#
to#minimize#a#loss#funcBon#(including#taking#logs#and#
manipulaBng#exponenBal#funcBons)#
3. Be#very#comfortable#converBng#from#least#squares#to#
the#normal#equaBons#(ideally,#you#can#easily#do#this#
with#an#added#L2#regularizaBon#term)#
4. Know#how#to#manipulate#expectaBons#(mean,#
variance,#biased,#risk)#
5. Know#the#effect#of#parameters/hyperTparameters#on#
algorithms#(what#happens#when#lambda/k/beta/etc#
goes#to#infinity)#
Advice#
6. Know#the#differences/tradeToffs#between#different#techniques#within#a#topic#(SVM#vs#logisBc#regression,#
kTmeans#vs#PCA)#
7. Review#decision#theory.#It#was#a#sizable#part#of#the#course,#but#didn’t#make#it#into#the#midterm.#
True/False#
c) The#maximum#likelihood#esBmator#for#the#parameter#θ#
of#a#uniform#distribuBons#[0,#θ]#is#unbiased.#
#False##
f) There#exists#a#oneTtoTone#feature#mapping#φ#for#every#
valid#kernel#
#False##
g) For#highTdimensional#data,#kTd#trees#can#be#slower#than#
brute#force#nearest#neighbor#search.#
#True##
h) If#we#had#infinite#data#and#infinitely#fast#computers,#
kNN#would#be#the#only#algorithm#we#would#study#in#189#
#True#
MulBple#Choice#
a) You#had#a#very#good#score#on#the#Kaggle#public#test#set,#but#did#poorly#on#the#private#test#set.#This#is#likely#
because#you#overfihed#by#submiQng#mulBple#Bmes#and#
changing#the#following#between#submissions:#
! λ,#your#penalty#term#
! η,#your#step#size#! ε,#your#convergence#criterion#! Fixing#a#random#bug#
MulBple#Choice#
c) PuQng#a#standard#Gaussian#prior#on#the#weights#for#
linear#regression#(w#~#N(0,1))#will#result#in#what#type#of#
posterior#distribuBon#on#the#weights?#
! Laplace#! Poisson#! Uniform#
! None#of#the#above#
MulBple#Choice#
e) Which#of#these#classifiers#could#have#generated#this#
decision#boundary?#
##
! Linear#SVM#
! LogisBc#Regression#! 1TNN#! None#of#the#above#
MulBple#Choice#
e) Which#of#these#classifiers#could#have#generated#this#
decision#boundary?#
##
! Linear#SVM#
! LogisBc#Regression#! 1TNN#! None#of#the#above#
MulBple#Choice#
e) Which#of#these#classifiers#could#have#generated#this#
decision#boundary?#
##
! Linear#SVM#
! LogisBc#Regression#! 1TNN#! None#of#the#above#
b) Is#there#a#relaBonship#between#this#type#of#input#perturbaBon#and#some#type#of#regularizer?#
#Yes,#L2#regularizer#