Commodity Machine Learning - (Andreas Mueller) Recent years have seen a widespread adoption of machine learning in industry and academia, impacting diverse areas from advertisement to personal medicine. As more and more areas adopt machine learning and data science techniques, the question arises on how much expertise is needed to successfully apply machine learning, data science and statistics. Not every company can afford a data science team, and getting your PhD in biology, no-one can expect you to have PhD-level expertise in computer science and statistics. This talk will summarize recent progress in automating machine learning and give an overview of the tools currently available. It will also point out areas where the ecosystem needs to improve in order to allow a wider access to inference using data science techniques. Finally we will point out some open problems regarding assumptions, and limitations of what can be automated. Andreas is an Research Engineer at the NYU Center for Data Science, building open source software for data science. Previously, he worked as a Machine Learning Scientist at Amazon, developing solutions for computer vision and forecasting problems. He is one of the core developers of the scikit-learn machine learning library, and has co-maintained it for several years. His mission is to create open tools to lower the barrier of entry for machine learning applications, promote reproducible science and democratize the access to high-quality machine learning algorithms.

1. amueller@nyu.edu
@amuellerml
@amueller
Amueller.io
Commodity
Machine Learning
and beyond
Andreas Müller
NYU, scikit-learn

2. To Apply Machine Learning!

3. What ML can do for you

4. Hi Andy,
I just received an email from the first tutorial
speaker, presenting right before you, saying
he's ill and won't be able to make it.
I know you have already committed yourself to
two presentations, but is there anyway you
could increase your tutorial time slot, maybe
just offer time to try out what you've taught?
Otherwise I have to do some kind of modern
dance interpretation of Python in data :-)
-LeahHi Andreas,
I am very interested in your Machine Learning
background. I work for X Recruiting who have
been engaged by Z, a worldwide leading supplier
of Y. We are expanding the core engineering
team and we are looking for really passionate
engineers who want to create their own story and
help millions of people.
Can we find a time for a call to chat for a few
minutes about this?
Thanks
Classification

5. Hi Andy,
I just received an email from the first tutorial
speaker, presenting right before you, saying
he's ill and won't be able to make it.
I know you have already committed yourself to
two presentations, but is there anyway you
could increase your tutorial time slot, maybe
just offer time to try out what you've taught?
Otherwise I have to do some kind of modern
dance interpretation of Python in data :-)
-LeahHi Andreas,
I am very interested in your Machine Learning
background. I work for X Recruiting who have
been engaged by Z, a worldwide leading supplier
of Y. We are expanding the core engineering
team and we are looking for really passionate
engineers who want to create their own story and
help millions of people.
Can we find a time for a call to chat for a few
minutes about this?
Thanks
Classification

6. Classification

7. Recommendations

8. Ranking

9. Applying machine learning is easy.

10. Applying machine learning is easy.
But it should be easier!

15. 2750+ research papers

16. from sklearn.naive_bayes import MultinomialNB
from sklearn.feature_extraction.text import CountVectorizer
from pipeline import make_pipeline
spam_classifier = make_pipeline(CountVectorizer(),
MultinomialNB())
spam_classifier.fit(email_texts, is_spam)
spam_classifier.predict(new_emails)
Fully Functional Spam Classifier

17. “The scikit-learn tutorials / documentation is so
good, one doesn't need a textbook anymore to
learn a new machine learning method.”

18. This is not enough!

19. Data size
Automation /
Expertise needed

20. Data size
Automation /
Expertise needed
Fits in Ram Single Machine Infinitely scalable
Library
One Click

21. Data size
Automation /
Expertise needed
Fits in Ram Single Machine Infinitely scalable
Library
One Click
Azure ML
Amazon Machine Learning
Skll

22. A single machine is (usually) enough.

24. We need open box methods.

25. We need black-box methods.

26. predict

27. Hyperparameter Optimization
Spearmint
Hyperopt
SMAC
… scikit-learn

28. From Eric Brochu, Vlad M. Cora and Nando de Freitas
Bayesian Optimization

29. Meta-Learning
optimization
Algorithm + ParametersDataset 1

30. Meta-Learning
optimization
Algorithm + Parameters
Dataset 3
optimization
Algorithm + Parameters
Dataset 2
optimization
Algorithm + Parameters
Dataset 1

31. Meta-Learning
Meta-Features 1
optimization
Algorithm + Parameters
Dataset 3
optimization
Algorithm + Parameters
Dataset 2
optimization
Algorithm + Parameters
Dataset 1
Meta-Features 2
Meta-Features 3
ML model

32. Meta-Learning
Meta-Features 1
optimization
Algorithm + Parameters
Dataset 3
optimization
Algorithm + Parameters
Dataset 2
optimization
Algorithm + Parameters
Dataset 1
Meta-Features 2
Meta-Features 3
ML model
New Dataset ML model Algorithm + Parameters

33. Data size
Automation /
Expertise needed
Fits in Ram Single Machine Infinitely scalable
Library
One Click
Skll Azure ML
Amazon Machine Learning

34. Data size
Automation /
Expertise needed
Fits in Ram Single Machine Infinitely scalable
Library
One Click
Skll Azure ML
Amazon Machine Learning

35. Data size
Automation /
Expertise needed
Fits in Ram Single Machine Infinitely scalable
Library
One Click
Skll
Next Stop
Azure ML
Amazon Machine Learning

36. 36
Release June 2016

37. 37
@amuellerml
@amueller
amueller@nyu.edu
http://amueller.io
Thank you.