Guy Blanc Neha Gupta Jane Lange Li-Yang Tan ! x f(x) 000010101 - - PowerPoint PPT Presentation

β–Ά
guy blanc neha gupta jane lange li yang tan
SMART_READER_LITE
LIVE PREVIEW

Guy Blanc Neha Gupta Jane Lange Li-Yang Tan ! x f(x) 000010101 - - PowerPoint PPT Presentation

Guy Blanc Neha Gupta Jane Lange Li-Yang Tan ! x f(x) 000010101 -1 " # 011011010 1 100100111 1 101001000 1 " -1 1 1 001010010 -1 1 1 ID3, C4.5, CART 1) Choose a good root $ [variable


slide-1
SLIDE 1

Guy Blanc Neha Gupta Jane Lange Li-Yang Tan

slide-2
SLIDE 2

x f(x) 000010101

  • 1

011011010 1 100100111 1 101001000 1 001010010

  • 1

𝑦! 𝑦" 𝑦#

  • 1

1 1 𝑦" 1 βˆ’1 ID3, C4.5, CART

slide-3
SLIDE 3

𝑦$

1) Choose a good root πΆπ‘£π‘—π‘šπ‘’(𝑔 𝑦$ < 0 ) πΆπ‘£π‘—π‘šπ‘’(𝑔 𝑦$ β‰₯ 0 ) 2) Recurse on subtrees β€œGood” root = relevant, important, influential [variable influence, information gain, Gini impurity]

slide-4
SLIDE 4

Noise-stable functions

[new in BGLT20]

Heuristics do poorly

[already well-known]

Monotone functions

[new in BLT20]

Heuristics do well New algorithm: provable guarantees for all functions [BGLT20]

slide-5
SLIDE 5

𝑦!

  • 1

1 1 π’šπŸ π’šπŸ‘ π’šπŸ’ π’šπŸ“ π’šπŸ” $2 $0 $5 $1200 $45 Query cost vector Randomized DT Stochastic node