Machine Learning runs predictive models that learn from existing data in order to forecast future behaviors, outcomes, and trends. A practical example for this is when you swipe your credit card somewhere and the bank verifies via Machine Learning if the transaction is likely to be a fraud. An other example are online shopping recommendations based on what you want to buy and what others have purchased before.
In this session, we'll lay the foundation for understanding the basics of Machine Learning, and see some practical examples of how it can be implemented on Azure
3. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Participate with us and stand a chance to win prizes!
Survey:
http://j.mp/abcsg-2016
Tweet Tag:
#GlobalAzure #GABSG
3
4. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Agenda
We’ll talk about:
• Machine Learning – What is it?
• Azure Machine Learning – The Basics
• Creating Models
• Using Models
• Demo – Predict HDB Resale Flat Price
4
We’ll ignore:
• Mathematical and statistical theories behind it all
• The Rise of the Machines
5. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Office Servers and Services (Office 365) MVP
Background: SharePoint + Office 365
Germany 6+ years in Singapore 3 years in Switzerland Back in Singapore!
Email : rene@modery.net
Twitter : @modery
Blog : http://modery.net
LinkedIn : https://www.linkedin.com/in/modery
Rene Modery
6. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
What is Machine Learning?
“Computing system that improve with experience”
Optimise a desired output or prediction using historical experiential data
6
Data
Output
Program
7. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Example - Recommendations
7
8. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Example – Office 365 Clutter
“Clutter learns from your actions to
determine the messages you are likely to
ignore. As less important messages
arrive, they are automatically moved to
the Clutter folder. Clutter does this by
leveraging Office Graph’s sophisticated
machine learning techniques to
determine which messages are Clutter. It
gets smarter over time, learning from
your prior actions with similar messages,
and assessing things like the type of
content and even how you are
addressed in the message. The Clutter
experience is personalized to each
individual and reflects an email
experience that adapts to your actions
and preferences without you having to do
anything.”
8
9. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Example – Fraudulent Credit Card Transactions
Date Vendor Location Amount Currency
14 December Thai Express Pte Ltd Singapore 32.95 SGD
9
15 December Singtel Pte Ltd Singapore 48.70 SGD
15 December Cold Storage Pte Ltd Singapore 43.40 SGD
16 December 7-Eleven Singapore 5.40 SGD
17 December Golden Village Pte Ltd Singapore 19.00 SGD
17 December Winter Time Pte Ltd Singapore 232.50 SGD
20 December Macy’s Herald Square, New York New York, USA 369.00 USD
Which of the following transactions is a fraud case? (Based on real events)
None, all were real! Last transaction looked fraudulent, bank investigation followed
Machine Learning doesn’t provide definite facts but predictions
10. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Machine Learning in Azure
Create
Model
Evaluate
Model
Deploy
Model
Test/Use
Model
10
DATA
Analyse
Cleanse
Feedback
Loop
Feedback
Loop
11. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Azure Machine Learning algorithms
11
Regression:
Predict one or
more continuous
variables, based
on other
attributes in the
dataset
Classification:
Classify data into
different
categories that
can then be
used to predict
one or more
discrete
variables, based
on other
attributes in the
dataset
12. Tweet: #GlobalAzure | http://singapore.azurebootcamp.net/
Put your Model to (good) Use
• Published Model available via REST API
for single or batch requests
• Create your own applications (web app,
desktop, mobile, …) that consume it
• Publish your final Model in the “Cortana
Intelligence Gallery”
• Publish your ML Web Service in the
Azure Marketplace
12