Ever wanted to do machine learning but don't know how to start? Join in me in a journey discovering ML and show you ML.NET Model Builder.
You can find an older version of the talk here: https://www.youtube.com/watch?v=ie8kQ5YU8eU
If you are interested in a full-day workshop, I'll have a workshop on 13th October 2020 at NDC Sydney.
https://ndcsydney.com/agenda/getting-started-with-machine-learning-using-mlnet-0wqu/0dtnq4bd90w
Simplified Machine Learning for Developers with ML.NET (Microsoft Reactor Abu Dhabi)
1. Microsoft Reactor Abu Dhabi
Machine Learning
Simplified for Developers with
ML.NET
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
V13.3
2. SSW Senior Software Architect
Jernej Kavka (JK)
@Jernej_kavka
github.com/jernejk
jkdev.me
linkedin.com/in/jernejkavka/
Brisbane AI User Group
.NET Core and EF Core dev by day
Cognitive Services and ML․NET enthusiast at night 🦸♂️
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
14. ML.NET
• MS machine learning SDK that works offline
• Simple yet powerful
• Already powering Microsoft products
• Support major ML models
• TensorFlow
• ONNX
• Awesome samples on GitHub
• https://github.com/dotnet/machinelearning-samples
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
15. My goal
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
POS AUTHORISATION DOTNETFOUNDATION
ORG REDMOND WA Card Used 0082
COFFEE 7 Melbourne
Audible Australia Melbourne
Investment
Food & Drink
Education
17. What is Machine Learning?
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
18. Machine Learning process
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
Improve
Data
Training data Model
ML Training
Desktop and mobile
Web
IoT
Feedback
Deploy
Usage
23. How do I start?
• Start with a scenario
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
24. Scenarios
• Is Tweet positive?
Binary classification
• Categorize bank transactions
Multi-class classification
• House price predictions
Regression
• Product recommendation
Clustering
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
25. Caveats before starting
• ML won’t magically solve your problem
• Understand your data
• Guide ML toward right solution
• Start with something simple like classification
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
29. Build model (acquire knowledge)
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
Training data Model
ML Training
30. Build model (acquire knowledge)
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
Training data Model
ML Training
31. ML.NET Model Builder
• Wizard built on top of ML.NET SDK
• Picks best trainer for given scenario and data
• Can be different one as the data changes over time
• Allows quick prototyping
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
35. Let’s do it
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka
POS AUTHORISATION DOTNETFOUNDATION
ORG REDMOND WA Card Used 0082
COFFEE 7 Melbourne
Audible Australia Melbourne
Investment
Food & Drink
Education
54. Summary
• Import CSV file
• Tweak columns to get better results
• Generate project with ML
• Tweaking generated code
• Minor issues with small datasets (100+ rows)
Join the Conversation @MSFTReactor #ReactorAbuDhabi @jernej_kavka