Lecture 1 Introductionx

wonderfuldistinctΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 3 χρόνια και 11 μήνες)

67 εμφανίσεις

Introduction

Lecture 1

METEO 474

Spring 2012



Provide
an accurate quantitative description of the
current and future weather


Do so without human intervention


Lower labor costs


Earlier forecasts


Higher number of forecasts


Use objective procedures via a computer: called variously


Statistical Forecasting


Pattern Recognition


Data Mining


Machine Learning


Artificial Intelligence

Objective Decision Making

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

2


Data


Observations as from METAR


KUNV 011515Z 00000KT 5SM BR CLR 17/17 A3017 RMK
AO2 LTG DSNT SW


Information


Gleaned from patterns in data as from contoured
maps



Data vs. Information

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

3

Information from Data

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

4


Overwhelmed by amount of available data


Amount of data doubles every 20 months!


Need
automated

way via computer to extract
information from data


Identify patterns in data and use them to help make
predictions


Data mining
in the field of
machine learning
is
the process by which we can find these patterns
objectively from a dataset


Learning ways to do this is focus of course

Data Mining Helps Extract Information
from Data

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

5


Goal
: Provide an accurate quantitative
description of the current and future weather


Forecast verification and valuation


Assess whether description of future weather is
usable and useful


Objective methodology discussed in class:


Decision Trees


Neural Nets

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Use an Automated Analysis
and
Forecasting System

6

Decision Tree: An Example


Decision Tree


A means for specifying a sequence of decisions
that need to be made &


The resulting recommendation


Example
:


Given the weather outlook and observations,
should a game be played?


Decision is “yes” or “no”

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

7

Outlook

Temperature

Humidity

Windy

Play

Sunny

Hot

High

False

no

Sunny

Hot

High

True

no

Overcast

Hot

High

False

yes

Rainy

Mild

High

False

yes

Rainy

Cool

Normal

False

yes

Rainy

Cool

Normal

True

no

Overcast

Cool

Normal

True

yes

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Weather Dataset from
Data Mining

Four Attributes and an Outcome to Classify Case as Play or Not Play

Etc.

3 possibilities 3 possibilities 2 possibilities 2 possibilities = 36 combinations

8


Develop a set of
classification rules
: They predict
the classification of the case as
play

or
not play


If outlook = sunny and humidity = high

then play = no


If outlook = rainy and windy = true

then play = no


If outlook = overcast





then play = yes


If humidity = normal





then play = yes


If none of the above




then play = yes


To make sense, the rules
must be applied in order


If not, then a wrong decision will be made as in:


If humidity = normal




then play = yes


Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Decision List

9


The need to apply in order leads to the “tree”


Decisions are reached at various levels in the list of
rules.


Produces a “tree” with “branches”


Example
: choosing a contact lens (soft, hard, none)


Series of questions developed from a set of data


Nested set of questions developed that are asked in the
correct order yielding a decision tree


Application with new cases leads to action
recommendation


Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Decision Tree

10

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Contact Lens Decision Tree

Source: Fig. 1.2 of
Data Mining, 3
rd

Ed.

11

Outlook

Temperature (F)

Humidity (%)

Windy

Play

Sunny

85

85

False

no

Sunny

80

90

True

no

Overcast

83

86

False

yes

Rainy

70

96

False

yes

Rainy

68

80

False

yes

Rainy

65

70

True

no

Overcast

64

65

True

yes

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Weather Dataset with Numeric Data

Etc.

If outlook = sunny and humidity > 83

then play = no

If outlook = sunny and humidity = high

then play = no

Mixed attribute problem: Classification rule more complex

12

Outlook

Temperature

Humidity

Windy

Play

Sunny

Hot

High

False

no

Sunny

Hot

High

True

no

Overcast

Hot

High

False

yes

Rainy

Mild

High

False

yes

Rainy

Cool

Normal

False

yes

Rainy

Cool

Normal

True

no

Overcast

Cool

Normal

True

yes

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Relationships within Weather Dataset

Regard all as attributes: How are they related?

Etc.

3 possibilities 3 possibilities 2 possibilities 2 possibilities = 36 combinations

13


Association rules
that predict any of the
attributes:


If temperature = cool then humidity = high


If outlook = sunny and play = no then
humidity = high


If windy = false and play = no then
outlook = sunny and humidity = high


These rules are 100% correct in this example,
but rarely will they be perfect predictors with
a more realistic case

Lecture 1 METEO 474 Sp 12 (Source: Prof. George Young)

Association Rules

14