Chasing the Pareto Frontier – In-Database Multi-Criteria Optimization with Skyline Queries

14
© 2014 EXASOL AG Chasing the Pareto Frontier In-Database Multi-Criteria Optimization with Skyline Queries

Transcript of Chasing the Pareto Frontier – In-Database Multi-Criteria Optimization with Skyline Queries

© 2014 EXASOL AG

1

Chasing the Pareto Frontier In-Database Multi-Criteria Optimization with Skyline Queries

© 2014 EXASOL AG

2

How to achieve optimal decisions

How to find optimal trade-offs for your criteria

Price/performance, costs/benefits, profit/risk, …

How to answer your essential questions

Which offering satisfies my customers needs but maximizes my margin?

Which campaigns reach my possible customers but minimize my efforts?

Skyline: Current research with University of Augsburg

© 2014 EXASOL AG

3

Example: Fund selection

Which are the best funds?

© 2014 EXASOL AG

4

Find optima in the presence of trade-offs between conflicting objectives

The problem

© 2014 EXASOL AG

5

Theory

Multi-dimensional optimization

In general, there is no single optimum

Performance

Risk

Better than

Which is better?

© 2014 EXASOL AG

6

Sorting? Filtering? Metrics? Scoring?

© 2014 EXASOL AG

7

Skyline – Multi-criteria optimization

Set of non-dominated elements

So-called Skyline or Pareto Frontier

Performance

Risk

Element’s dominated area

© 2014 EXASOL AG

8

The Power of Skyline Queries

Exclude the sub-optimal values

© 2014 EXASOL AG

9

Skyline SQL example

Evaluate the ‘optimal’ holiday trip

duration should be as high as possible

price should be as low as possible

Returns a set of optimal trips per day

SELECT *

FROM trips

WHERE country='USA'

PREFERRING HIGH duration

PLUS LOW price

PARTITION BY start_day;

© 2014 EXASOL AG

10

Live Demo

Public available dataset:

Renewable energy map

EXASolo Community Edition

EXAplus SQL Client

© 2014 EXASOL AG

11

How does it scale ? Benchmark Results

http://www.btw-2015.de/res/proceedings/Hauptband/Ind/Mandl-Preference_Analytics_in_EXAS.pdf

© 2014 EXASOL AG

12

Summary

Given objective, multidimensional criteria, Skyline queries:

compute Pareto Frontier / non-dominated solutions

reduce huge data volumes to the relevant subset

accelerate and optimize your decision making process

Try out yourself:

https://www.exasol.com/support/browse/SOL-104 What is Skyline? + Link to Videos

https://www.exasol.com/support/browse/SOL-240 Skyline Tutorial: Energymap

© 2014 EXASOL AG

13

EXASolution - Powerful Analytic Engine

Query Interfaces • SQL • ODBC • JDBC

In-database Analytics • Java, Python, R, Lua • Geo-spatial • Skyline

100% built from the scratch

Connectors • Hadoop et al. • 3rd party JDBC • Native Oracle

In-memory - Columnar – Massively Parallel Processing

© 2014 EXASOL AG

14

Questions ?

More details and a FREE community version of our database available at

www.exasol.com

Email:

[email protected]

Twitter : @EXASOLAG