8. What is ML
▪ “Machine learning systems automatically learn
programs from data” Domingos, 2012
Computation
Data
Cat
Program
ML Approach
Data
Result
9. What is ML
▪ “Machine learning systems automatically learn
programs from data” Domingos, 2012
10. What is ML
▪ “Machine learning systems automatically learn
programs from data” Domingos, 2012
▪ Coursera
11. What is ML
▪ “Machine learning systems automatically learn
programs from data” Domingos, 2012
▪ Coursera
▪ Kaggle
12. Your Problem
▪ Some examples from Kaggle:
▪ Which driver will file insurance in next 12 months?
kaggle.com
13. Your Problem
▪ Some examples from Kaggle:
▪ Which driver will file insurance in next 12 months?
▪ Which content will be clicked by a user?
kaggle.com
14. Your Problem
▪ Some examples from Kaggle:
▪ Which driver will file insurance in next 12 months?
▪ Which content will be clicked by a user?
▪ Which article will be bought by a customer in the next 12 months?
kaggle.com
24. Models and Evaluation
1. Come up with a baseline
▪ Naive approach
▪ Something you can do without ML
(mean, median)
25. Models and Evaluation
1. Come up with a baseline
▪ Naive approach
▪ Something you can do without ML
(mean, median)
2. Start with a simple model
▪ “Draw the line, yes this counts as
machine learning”
▪ Try to be better than the naive
approach
26. Models and Evaluation
1. Come up with a baseline
▪ Naive approach
▪ Something you can do without ML
(mean, median)
2. Start with a simple model
▪ “Draw the line, yes this counts as
machine learning”
▪ Try to be better than the naive
approach
3. Build a more complex model
▪ Beat the simple model
▪ Iterate
27. Models and Evaluation
1. Come up with a baseline
▪ Naive approach
▪ Something you can do without ML
(mean, median)
2. Start with a simple model
▪ “Draw the line, yes this counts as
machine learning”
▪ Try to be better than the naive
approach
3. Build a more complex model
▪ Beat the simple model
▪ Iterate
spotify.com
28. Models and Evaluation
1. Come up with a baseline
▪ Naive approach
▪ Something you can do without ML
(mean, median)
2. Start with a simple model
▪ “Draw the line, yes this counts as
machine learning”
▪ Try to be better than the naive
approach
3. Build a more complex model
▪ Beat the simple model
▪ Iterate
▪ Ideally stop when effort > gain
spotify.com