Machine Learning

Data

Jeff Howbert Introduction to Machine Learning Winter 2012 1

 Data topics

z Types of attributes z Types of datasets

z Data quality issues z Preprocessing

z Transformations z Summary statistics

z Visualization

Jeff Howbert Introduction to Machine Learning Winter 2012 2

 What is data?

z Collection of data objects and Attributes

their attributes

z An attribute is a property or Tid Refund Marital Taxable

Status Income Cheat
characteristic of an object
1 Yes Single
g 125K No
E
Examples:
l eye color
l off a
2 No Married 100K No
person, temperature, etc.
3 No Single 70K No
Attribute is also known as
4 Yes Married 120K No
variable field
variable, field, characteristic
characteristic,
5 No Divorced 95K Yes
or feature Objects
6 No Married 60K No
z A collection of attributes
7 Yes Divorced 220K No
describe an object
8 No Single 85K Yes
Object is also known as
9 No Married 75K No
record, point, case, sample,
10 No Single 90K Yes
y, or instance
entity, 10

Jeff Howbert Introduction to Machine Learning Winter 2012 3

 Attribute values

z Attribute values are numbers or symbols assigned to an

attribute

z Distinction between attributes and attribute values

Same attribute can be mapped
pp to different attribute
values
Example: height can be measured in feet or meters

Different attributes can be mapped to the same set of

values
Example: Attribute values for ID and age are integers
But properties of attribute values can be different

ID has no limit but age has a maximum and minimum value

Jeff Howbert Introduction to Machine Learning Winter 2012 4

 Types of attributes

z There are different types of attributes

Nominal
Examples: ID numbers, eye color, zip codes
Ordinal
Examples: rankings (e.g., taste of potato chips on a scale
from 1-10), grades, height in {tall, medium, short}
Interval
Examples: calendar dates, temperatures in Celsius or
Fahrenheit.
Ratio
R ti
Examples: temperature in Kelvin, length, time, counts

Jeff Howbert Introduction to Machine Learning Winter 2012 5

 Properties of attributes

z The type of an attribute depends on which of the following

properties it possesses:
Distinctness: =
Order: < >
Addition: + -
Multiplication: */

Nominal attribute: distinctness

Ordinal attribute: distinctness & order
Interval attribute: distinctness, order & addition
Ratio attribute: all four properties

Jeff Howbert Introduction to Machine Learning Winter 2012 6

Attribute Description Examples Statistical
Type Operations

Nominal The values of a nominal attribute are zip codes, employee mode, entropy,
just different names, i.e., nominal ID numbers, eye color, contingency
attributes provide only enough sex: {male, female} correlation, 2 test
information to distinguish one object
f
from another.
h (=,( ))

Ordinal The values of an ordinal attribute hardness of minerals, median, percentiles,

provide enough information to order {good, better, best}, rank correlation,
objects (<,
objects. (< >) grades street numbers
grades, run tests
tests, sign tests

Interval For interval attributes, the calendar dates, mean, standard

diff
differences between
b values
l are temperature in
i Celsius
C li d i i Pearson's
deviation, P '
meaningful, i.e., a unit of or Fahrenheit correlation, t and F
measurement exists. tests
(+, - )

Ratio For ratio variables, both differences temperature in Kelvin, geometric mean,
and ratios are meaningful. (*, /) monetary quantities, harmonic mean,
counts, age, mass, percent variation
length, electrical
current
Attribute Allowed Transformations Comments
Level

Nominal Any permutation of values If all employee ID numbers

were reassigned, would it
make any difference?

Ordinal An order preserving change of An attribute encompassing

values ii.e.,
values, e the notion of {good,
{good better
new_value = f(old_value) best} can be represented
where f is a monotonic function. equally well by values
{1, 2, 3} or {0.5, 1, 10}.

Interval new_value =a * old_value + b The Fahrenheit and Celsius

where a and b are constants temperature scales differ in
terms of where their zero
value is and the size of a
unit (degree).

R ti
Ratio new_value
l = a * old_value
ld l Length
L th can be
b measuredd in
i
meters or feet.
 Discrete and continuous attributes

z Discrete attribute
Has only a finite or countably infinite set of values
Examples: zip codes, counts, or the set of words in a collection of
documents
Often represented as integer variables.
Note: binary attributes are a special case of discrete attributes

z Continuous attribute
Has real numbers as attribute values
Examples: temperature, height, or weight.
Practically,
Practically real values can only be measured and represented
using a finite number of digits.
Continuous attributes are typically represented as floating-point
variables.

Jeff Howbert Introduction to Machine Learning Winter 2012 9

 Types of data sets

z Record
Data matrix
Document data
Transaction data
z G h
Graph
World Wide Web
Molecular structures
z Ordered
Spatial data
Temporal (time series) data
Sequential data
Genetic
G ti sequence data
d t

Jeff Howbert Introduction to Machine Learning Winter 2012 10

 Record data

z Data that consists of a collection of records, each

of which consists of a fixed set of attributes
Tid Refund Marital Taxable
Status Income Cheat

1 Yes Single 125K No

2 No Married 100K No
3 No Single 70K No
4 Yes Married 120K No
5 No Divorced 95K Yes
6 No Married 60K No
7 Yes Divorced 220K No
8 No Single 85K Yes
9 No Married 75K No
10 No Single 90K Yes
10

Jeff Howbert Introduction to Machine Learning Winter 2012 11

 Data matrix

z If data objects have the same fixed set of numeric

attributes, then the data objects can be thought of as
points in a multi-dimensional space, where each
dimension represents a distinct attribute.

z Such data set can be represented by an m x n matrix,

where there are m rows, one for each object, and n
columns,
l one ffor each
h attribute
tt ib t

Projection Projection Distance Load Thickness

of x Load of y load

10.23 5.27 15.22 2.7 1.2

12 65
12.65 6 25
6.25 16 22
16.22 22
2.2 11
1.1

Jeff Howbert Introduction to Machine Learning Winter 2012 12

 Document data

z Each document becomes a `term' vector,

each term is a component (attribute) of the vector,
the value of each component is the number of times the
corresponding term occurs in the document.

timeou

seaso
coach

game
score
team

ball

lost
pla

wi
n
y
m

on
e

e
h

ut
Jeff Howbert Introduction to Machine Learning Winter 2012 13
 Transaction data

z A special type of record data, where

Each record (transaction) involves a set of items.
For example, consider a grocery store. The set of products
purchased by a customer during one shopping trip constitute a
transaction, while the individual products that were purchased
are the items.

TID Items
1 Bread, Coke, Milk
2 Beer, Bread
3 Beer,
ee , Coke,
Co e, Diaper,
pe , Milk
4 Beer, Bread, Diaper, Milk
5 Coke, Diaper, Milk

Jeff Howbert Introduction to Machine Learning Winter 2012 14

 Graph data

z Examples: Generic graph and HTML Links

<a href="papers/papers.html#bbbb">
Data Mining </a>
<li>
2 <a href="papers/papers.html#aaaa">
Graph Partitioning </a>
<li>
5 1 <a href="papers/papers.html#aaaa">
Parallel Solution of Sparse Linear System of Equations </a>
<li>
2 <a href="papers/papers.html#ffff">
N-Body Computation and Dense Linear System Solvers
5

Jeff Howbert Introduction to Machine Learning Winter 2012 15

 Chemical data

z Benzene molecule: C6H6

Jeff Howbert Introduction to Machine Learning Winter 2012 16

 Ordered data

z Sequences of transactions
It
Items/Events
/E t

An element of
the sequence

Jeff Howbert Introduction to Machine Learning Winter 2012 17

 Ordered data

z Genomic sequence data

GGTTCCGCCTTCAGCCCCGCGCC
CGCAGGGCCCGCCCCGCGCCGTC
GAGAAGGGCCCGCCTGGCGGGCG
GGGGGAGGCGGGGCCGCCCGAGC
CCAACCGAGTCCGACCAGGTGCC
CCCTCTGCTCGGCCTAGACCTGA
GCTCATTAGGCGGCAGCGGACAG
GCCAAGTAGAACACGCGAAGCGC
TGGGCTGCCTGCTGCGACCAGGG

Jeff Howbert Introduction to Machine Learning Winter 2012 18

 Ordered data

z Spatio-temporal data

Average monthly
temperature of
land and ocean

Jeff Howbert Introduction to Machine Learning Winter 2012 19

 Stages of knowledge extraction

Interpretation /
evaluation

Knowledge
Machine
learning

Transformation Patterns

Preprocessing
Transformed
data

Selection Preprocessed
d t
data
Data Target
data

Jeff Howbert Introduction to Machine Learning Winter 2012 20

 Data quality

z What kinds of data quality problems?

z How can we detect problems with the data?
z What can we do about these problems?

z Examples of data quality problems:

noise and outliers
missing values
duplicate data

Jeff Howbert Introduction to Machine Learning Winter 2012 21

 Noise

z Noise refers to random modification of original values

z Examples:
distortion of a persons voice when talking on a poor phone
snow on television screen

Two sine waves Two sine waves + noise

Jeff Howbert Introduction to Machine Learning Winter 2012 22

 Noise

z Dealing with noise

Mostly you have to live with it
Certain kinds of smoothing or averaging can be
helpful
In the right domain (e.g. signal processing),
transformation to a different space can get rid of
majority of noise

Jeff Howbert Introduction to Machine Learning Winter 2012 23

 Outliers

z Outliers are data objects with characteristics that

are considerably different than most of the other
data objects in the data set

Jeff Howbert Introduction to Machine Learning Winter 2012 24

 Outliers

z Dealing with outliers

There are robust statistical methods for detecting
outliers
In some situations, you want to get rid of outliers
but be judicious they may carry useful, even important
information
In other situations,
situations the outliers are the objects of
interest
anomaly detection

Jeff Howbert Introduction to Machine Learning Winter 2012 25

 Missing values

z Reasons for missing values

Information is not collected
(e.g., people decline to give their age and weight)
Attributes may not be applicable to all cases
(e g annual income is not applicable to children)
(e.g.,

z Handling
g missing
g values
Eliminate data objects
Estimate missing values
Ignore the missing value during analysis
Replace with all possible values (weighted by their
probabilities)

Jeff Howbert Introduction to Machine Learning Winter 2012 26

 Duplicate data

z Data set may include data objects that are

duplicates, or almost duplicates of one another
Major issue when merging data from heterogeous
sources

z Example:
Same person with multiple email addresses

z Data cleaning
Includes process of dealing with duplicate data issues

Jeff Howbert Introduction to Machine Learning Winter 2012 27

 Data preprocessing

z Aggregation
z Sampling
z Discretization and binarization
z Attribute transformation
z Feature creation
z Feature
F t selection
l ti
Choose subset of existing features
z Di
Dimensionality
i lit reduction
d ti
Create smaller number of new features through linear
or nonlinear combination of existing features

Jeff Howbert Introduction to Machine Learning Winter 2012 28

 Aggregation

z Combining two or more attributes (or objects) into

a single attribute (or object)

z Purpose
Data reduction
Reduce the number of attributes or objects
Change of scale
Cities aggregated into regions, states, countries, etc
More stable data
Aggregated data tends to have less variability

Jeff Howbert Introduction to Machine Learning Winter 2012 29

 Aggregation

Variation of precipitation in Australia

Standard
S a da d de
deviation
a o o of a
average
e age Standard
S a da d de
deviation
a o o of a
average
e age
monthly precipitation yearly precipitation

Jeff Howbert Introduction to Machine Learning Winter 2012 30

 Sampling

z Sampling is the main technique employed for data

selection.
Often used for both preliminary investigation of the data and the
final data analysis.

z Statisticians sample because obtaining the entire set of

data of interest is too expensive or time consuming.

z Sampling is used in data mining because processing the

entire
ti sett off data
d t off interest
i t t is
i too
t expensive
i or time
ti
consuming.

Jeff Howbert Introduction to Machine Learning Winter 2012 31

 Sampling

z The key principle for effective sampling is the following:

Using a sample will work almost as well as using the entire data
set, provided the sample is representative.
A sample is representative if it has approximately the same
distribution of properties (of interest) as the original set of data

Jeff Howbert Introduction to Machine Learning Winter 2012 32

 Types of sampling

z Simple random sampling

There is an equal probability of selecting any particular item.
z Sampling without replacement
As each item is selected, it is removed from the population.
z Sampling with replacement
Objects are not removed from the population as they are
selected for the sample.
Same object can be selected more than once
once.
z Stratified sampling
Split the data into several partitions; then draw random samples
from each partition
partition.
Example: During polling, you might want equal numbers of male
and female respondents. You create separate pools of men and
women and sample separately from each
women, each.

Jeff Howbert Introduction to Machine Learning Winter 2012 33

 Sample size

8000 points 2000 Points 500 Points

Jeff Howbert Introduction to Machine Learning Winter 2012 34

 Sample size
z What sample size is necessary to get at least one object
from each of 10 equal-sized groups?

Jeff Howbert Introduction to Machine Learning Winter 2012 35

Discretization without using class labels

Data Equal interval width

Equal frequency k-means

Jeff Howbert Introduction to Machine Learning Winter 2012 36

 Discretization using class labels

z Entropy based approach

3 categories for both x and y 5 categories for both x and y

Jeff Howbert Introduction to Machine Learning Winter 2012 37

 Attribute transformation

Definition:
A function that maps the entire set of
values of a given attribute to a new set
of replacement values, such that each
old value can be identified with one of
the new values.

Jeff Howbert Introduction to Machine Learning Winter 2012 38

 Attribute transformation

z Simple functions
Examples of transform functions:
xk log( x ) ex |x|
Often used to make the data more like some standard distribution,
to better satisfy assumptions of a particular algorithm.
Example: discriminant analysis explicitly models each class distribution as a
multivariate Gaussian

log( x )

Jeff Howbert Introduction to Machine Learning Winter 2012 39

 Attribute transformation

z Standardization or normalization
Usually involves making attribute:
mean =0
standard deviation =1
in MATLAB, use zscore() function
Important when working in Euclidean space and attributes have
very different numeric scales.
Also necessaryy to satisfy
y assumptions
p of certain algorithms.
g
Example: principal component analysis (PCA) requires each attribute to be
mean-centered (i.e. have mean subtracted from each value)

Jeff Howbert Introduction to Machine Learning Winter 2012 40

 Transform data to a new space

z Fourier transform
Eliminates noise present in time domain

Two sine waves Two sine waves + noise Frequency

Jeff Howbert Introduction to Machine Learning Winter 2012 41

 Summary statistics and visualization

Lets
Let s use a tool thats
that s good at those things

PowerPoint isn
isntt it

Jeff Howbert Introduction to Machine Learning Winter 2012 42

 Iris dataset

z Many exploratory data techniques are nicely

illustrated with the iris dataset.
Dataset created by famous statistician Ronald Fisher
150 samples of three species in genus Iris (50 each)
Iris setosa
Iris versicolor
Iris virginica
g
Four attributes
sepal width
sepal length

petal width
Iris virginica. Robert H. Mohlenbrock. USDA
petal length NRCS. 1995. Northeast wetland flora: Field office
guide to plant species
species. Northeast National
Species is class label Technical Center, Chester, PA. Courtesy of USDA
NRCS Wetland Science Institute.
Jeff Howbert Introduction to Machine Learning Winter 2012 43