Slide 1

Slide 1 text

Better than Deep Learning: Gradient Boosting Machines (GBM) in R Szilárd Pafka, PhD Chief Scientist, Epoch (USA) eRum Conference, Budapest May 2018

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

Disclaimer: I am not representing my employer (Epoch) in this talk I cannot confirm nor deny if Epoch is using any of the methods, tools, results etc. mentioned in this talk

Slide 4

Slide 4 text

Source: Andrew Ng

Slide 5

Slide 5 text

Source: Andrew Ng

Slide 6

Slide 6 text

Source: Andrew Ng

Slide 7

Slide 7 text

No content

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

No content

Slide 11

Slide 11 text

Source: https://twitter.com/iamdevloper/

Slide 12

Slide 12 text

No content

Slide 13

Slide 13 text

...

Slide 14

Slide 14 text

No content

Slide 15

Slide 15 text

No content

Slide 16

Slide 16 text

http://www.cs.cornell.edu/~alexn/papers/empirical.icml06.pdf http://lowrank.net/nikos/pubs/empirical.pdf

Slide 17

Slide 17 text

http://www.cs.cornell.edu/~alexn/papers/empirical.icml06.pdf http://lowrank.net/nikos/pubs/empirical.pdf

Slide 18

Slide 18 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL

Slide 19

Slide 19 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends

Slide 20

Slide 20 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all

Slide 21

Slide 21 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning

Slide 22

Slide 22 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning / ensembles

Slide 23

Slide 23 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning / ensembles feature engineering

Slide 24

Slide 24 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning / ensembles feature engineering / other goals e.g. interpretability

Slide 25

Slide 25 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning / ensembles feature engineering / other goals e.g. interpretability the title of this talk was misguided

Slide 26

Slide 26 text

structured/tabular data: GBM (or RF) very small data: LR very large sparse data: LR with SGD (+L1/L2) images/videos, speech: DL it depends / try them all / hyperparam tuning / ensembles feature engineering / other goals e.g. interpretability the title of this talk was misguided but so is recently almost every use of the term AI

Slide 27

Slide 27 text

Source: Hastie etal, ESL 2ed

Slide 28

Slide 28 text

Source: Hastie etal, ESL 2ed

Slide 29

Slide 29 text

Source: Hastie etal, ESL 2ed

Slide 30

Slide 30 text

Source: Hastie etal, ESL 2ed

Slide 31

Slide 31 text

No content

Slide 32

Slide 32 text

No content

Slide 33

Slide 33 text

10x

Slide 34

Slide 34 text

No content

Slide 35

Slide 35 text

No content

Slide 36

Slide 36 text

10x

Slide 37

Slide 37 text

No content

Slide 38

Slide 38 text

No content

Slide 39

Slide 39 text

No content

Slide 40

Slide 40 text

No content

Slide 41

Slide 41 text

No content

Slide 42

Slide 42 text

No content

Slide 43

Slide 43 text

No content

Slide 44

Slide 44 text

No content

Slide 45

Slide 45 text

No content

Slide 46

Slide 46 text

No content

Slide 47

Slide 47 text

No content

Slide 48

Slide 48 text

http://www.jmlr.org/papers/volume13/bergstra12a/bergstra12a.pdf

Slide 49

Slide 49 text

http://www.argmin.net/2016/06/20/hypertuning/

Slide 50

Slide 50 text

No content

Slide 51

Slide 51 text

No content

Slide 52

Slide 52 text

No content

Slide 53

Slide 53 text

No content

Slide 54

Slide 54 text

No content

Slide 55

Slide 55 text

More:

Slide 56

Slide 56 text

No content

Slide 57

Slide 57 text

No content

Slide 58

Slide 58 text

Backup Slides

Slide 59

Slide 59 text

All benchmarks are wrong, but some are useful

Slide 60

Slide 60 text

No content

Slide 61

Slide 61 text

No content

Slide 62

Slide 62 text

No content

Slide 63

Slide 63 text

No content

Slide 64

Slide 64 text

No content

Slide 65

Slide 65 text

No content

Slide 66

Slide 66 text

No content