CSE217 INTRODUCTION TO DATA SCIENCE LECTURE 10: DATA ENGINEERING - - PowerPoint PPT Presentation

CSE217 INTRODUCTION TO DATA SCIENCE

Spring 2019 Marion Neumann

LECTURE 10: DATA ENGINEERING

RECAP: FEATURE ENGINEERING

2

5 Good Reasons for Feature Engineering: 1) get better represented features (scaling, standardization)

à improve model training and prediction quality

2) get more expressive features

à improve model training and prediction quality à ACTIVITY 1

3) get more representative features

à remove noise

4) get less features

à more efficient computation

5) represent features in 2d or 3d

à visualization

6) bring data in to vector representation

à not all data necessarily comes in vector form ⃑

" = "$ ⋮ "& What data does not come in vector form?

• data that does not come as numerical vectors

à requires feature extraction

Example: images

COMPLEX DATA

3

• data that does not come as numerical vectors

à requires feature extraction

Example: text

COMPLEX DATA

4 great small location friends …

Same great flavor and friendly service as in the S 18th street

• location. This location is not as small but it's hard to talk to friends.

Thankfully there is great outdoor seating to escape the noise.

• data that does not come as numerical vectors

à requires feature extraction

Example: street networks

COMPLEX DATA

5

• data that does not come as numerical vectors

à requires feature extraction

Example: chemical compounds

COMPLEX DATA

6 mutagenic non-mutagenic ???

• data that does not come as numerical vectors

à requires feature extraction

Example: point clouds

COMPLEX DATA

7

• data that does not come as numerical vectors

à requires feature extraction

Example: social networks

COMPLEX DATA

8

FEATURE EXTRACTION

• feature extraction is challenging
• which features to compute and use?
• too many choices and combinations
• NNs can be used to learn features
• sometimes those features are even interpretable
• works well for images and text

9

FEATURE EXTRACTION VS LEARNING

10

FEATURE EXTRACTION VS LEARNING

11

Same great flavor and friendly service as in the S 18th street location. This location is not as small but it's hard to talk to friends. Thankfully there is great

• utdoor seating to escape the noise.

Same great flavor and friendly service as in the S 18th street location. This location is not as small but it's hard to talk to friends. Thankfully there is great

• utdoor seating to escape the noise.

politics sports culture politics sports culture

FEATURE EXTRACTION VS LEARNING

12

politics sports culture Train NN on very large corpus simple ML model

• kNN
• random forest
• …

positive negative Train separate model

• n new (small) data

!" = 0.23 !( = 3.42 !*= 0.89

Use pre-trained NN

DATA ENGINEERING

• instead of removing or creating features
• create or remove data points!
• Why could that be useful?
• remove outliers
• create more training examples à data augmentation

13

CAUSES OF OUTLIERS

• data entry errors (human errors)
• measurement errors (instrument errors)
• experimental errors
• data extraction error
• experiment planning/executing errors
• intentional
• dummy outliers made to test detection methods
• data processing errors
• data manipulation
• data set unintended mutations
• sampling errors
• extracting or mixing data from wrong or various sources
• natural à not an error, novelties in data

14

OUTLIER DETECTION

• standardize your data
• empirical rule:

15

• utliers

91

• utliers
• 31

! = #$%#&%%#'

(

= )

( * +,) (

• +

Z-Score Analysis (parametric approach à makes the parametric assumption that the features are normally distributed)

samples

7 8 48 55 68

. = ./ =

) (0) ∑23$

'

#204 &

sample standard deviation sample mean

DATA AUGMENTATION

• easy for images

à preform image transformations

16

DATA AUGMENTATION

Why à NNs need a ton of training data

• NN performance is very sensitive to adversarially created noise
• collecting (and labelling data) is very time consuming/expensive

17

+ .007 ⇥ = x sign(rxJ(θ, x, y)) x + ✏ sign(rxJ(θ, x, y)) y =“panda” “nematode” “gibbon” w/ 57.7% confidence w/ 8.2% confidence w/ 99.3 % confidence

18

• [DSFS]
• Ch18: Neural Networks (p213-218)
• http://playground.tensorflow.org

SUMMARY & READING

• More expressive features (even simple transformations) can

greatly improve the training time and model quality.

• Feature learning is useful to deal with non-vectorial input

data.

• Data engineering can improve supervised models by removing
• utliers or data augmentation.
• Neural Networks are tricky to train and very sensitive to noise.