Machine learning

1. Machine Learning
Concepts and Applications

2. Eliézer Zarpelão
 Técnico em Informática - UNICAMP
 Graduação em Sistemas de Informação - USP
 Especialização em Engenharia de Software - Unicamp
 Mestrado (especial) - Unicamp
 Co-fundador - ZarpSystem
 Analista de Sistemas - Unicamp
 Professor Ensino Superior - IESCAMP
 Coordenador de Curso - IESCAMP
Processo nº 2016/50250-1

4. What is ML?
“Machine Learning is the science (and art) of programming computers so they can learn from data”.
[Aurélien Géron, 2017]
“Field of study that gives computers the ability to learn without being explicitly programmed.”
[Arthur Samuel, 1959]
“A computer program is said to learn from experience E with respect to some task T and some
performance measure P, if its performance on T, as measured by P, improves with experience E.”
[Tom Mitchell, 1997]

5. What is ML?
F = a * m

6. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

7. Sample Data
Dados estruturados Dados não estruturados
Petal Lenght Petal Width Sepal Lenght Sepal Width Species
5.2 3.3 1.3 0.3 0
4.7 1.4 3.1 0.2 0
... ... ... ... ...
6.4 3.2 5.2 2.0 3
5.2 3.3 1.3 0.3 3
5.9 3.0 5.1 1.9 4

8. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

9. Feature Engineering
Id Survived Pclass Name Sex Age SibSp Parch Ticket Fare Cabin
Embark
ed
1 0 3 Braund, Mr. Owen Harris male 22 1 0 A/5 21171 7,25 S
2 1 1 Cumings, Mrs. John Bradley (Florence Briggs Thayer) female 38 1 0 PC 17599 71,2833 C85 C
3 1 3 Heikkinen, Miss. Laina female 26 0 0
STON/O2.
3101282 7,925 S
4 1 1 Futrelle, Mrs. Jacques Heath (Lily May Peel) female 35 1 0 113803 53,1 C123 S
5 0 3 Allen, Mr. William Henry male 35 0 0 373450 8,05 S
6 0 3 Moran, Mr. James male 0 0 330877 8,4583 Q
7 0 1 McCarthy, Mr. Timothy J male 54 0 0 17463 51,8625 E46 S

10. Data preparation is
80% of machine
learning

11. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

12. ML algorithms

13. ML algorithms

14. ML algorithms

15. ML algorithms

16. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

17. Gradient Descent
Goal: Find best weight vector

18. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

19. Train Model
Underfitting o/ Overfitting

20. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

21. Test Model
o/ Overfitting
● Treinamento
● Teste

22. Test Model
Underfitting o/ Overfitting

23. Test Model
Underfitting:
● Erro no treinamento alto
● Erro no teste alto
Just right:
● Erro no treinamento baixo
● Erro no teste baixo
Overfitting:
● Erro no treinamento baixo
● Erro no teste alto

24. Machine Learning
Sample
Data
Feature
Selection
Choose
Algorithm
Train
Model
Results
OK?
Test
Model
Results
OK?
Use in real
world

25. Soccer Application
https://www.kaggle.com/kvnchn/notebook399529d6ab/data
 +25k matches
 +10k players
 11 European Countries with their lead championship
 Seasons 2008 to 2016
 Players and Teams' attributes
 Team line up with squad formation (X, Y coordinates)
 Detailed match events (goal types, possession, corner, cross, fouls, cards etc...) for +10k
matches

26. The secret of playing football: Brazil versus the Netherlands
Professors
Processo nº 2016/50250-1
J. W ainer
– Artificial Intelligence
– Medical Informatics
– Information and Society
S. Goldenstein
– Vision
– Computer Graphics
– Forensics
Retrieval
arning
databases
On Leave
mputing
ormatics
Ricardo
Torres
IC
UNICAMP
Claudio
Gobatto
FCA
UNICAMP
Sergio
Cunha
FEF
UNICAMP
Ricardo
Barros
FEF
UNICAMP
Luiz Barreto
Martins
FEF
UNICAMP
Felipe
Moura
UEL
Paulo
Santiago
USP
Ricardo Anido
Institute of
Computing
UNICAMP

27. The secret of playing football: Brazil versus the Netherlands
PhD Research: Ícaro Dourado
Title: Graph-based search
Processo nº 2016/50250-1

28. The secret of playing football: Brazil versus the Netherlands
PhD Research: Samuel Fadel
Title: Deep learning for temporal networks in prediction problems
Processo nº 2016/50250-1

29. Nome: Eliézer Zarpelão
Email: eliezer.zarpelao@gmail.com
Empreendedor apaixonado por desenvolvimento Web, desenvolvedor com
anos de experiência e co-fundador da ZarpSystem. Atualmente atua como
analista de sistemas na Unicamp, coordenador de curso e professor
universitário na Faculdade IESCAMP e violoncelista nas horas vagas.
Mestrando no IC/Unicamp, Especialista em Engenharia de Software pela
Unicamp, Graduado em Sistemas de Informação pela USP
Obrigado!