Download - An introduction to azure machine learning
An Introduction to Azure Machine
LearningDouglas M. Kline, Ph.D.
Professor of Information Systems, UNC WilmingtonDatabase by Doug
About Me• From Akron, OH• Professor of Information Systems• Teaching: Database, Software Development, others• Research: Neural Networks, Security, Pedagogy, Analytics, IT Strategy,
etc.• Professional: • DatabaseByDoug: SQL Server Consulting (internals, performance tuning)• DatabaseByDoug: https://www.youtube.com/c/databasebydoug• DatabaseByDoug: http://douglaskline.blogspot.com/
Overview• What’s Azure?• What’s Azure Machine Learning?• Getting Data• Model Building• Publishing as a Web Service• Consuming the Web Service• Conclusion
What’s Azure?• Microsoft’s cloud computing services platform• Storage, Bandwidth, Computing, services• Self-serve• Metered – pay for what you use• Helps to be aware of charges
What’s Azure Machine Learning?• Cloud service for analytics• Machine Learning Studio• Visual experiment designer, drag and drop• Pre-defined method blocks
• Classification, clustering, time series, prediction, statistics, etc.• Data input, output, transformations, etc.• Experiment control: data partitioning, model definition, training, scoring, evaluation, etc.
• R blocks
• Deploy models as Web Services• web service marketplace
Getting Data• Sources: SQL, Storage, CSV• Manipulation: SQL, column selection, sampling• Basic Stats• R block• Cache Data Set• Save Data Set
Demo: Get Data from Azure SQL• Input Block• Wizard• SELECT a sample – randomUniform• Visualizations• Summarize Block• Feature Selection
• Automatic• Interactive
• Save as Data Set• Simple R Block
Demo: Model Building• Import Data Set• Split Data: training / testing• Two 2-Group Classification Models:
• NN• Boosted Decision Tree
• Model training• Model scoring (training/testing)• Model evaluation
• Training vs. testing• Model A vs. Model B
• Recalibrate• Save Trained Model
Metrics• Accuracy – % correctly classified, positive or negative • Precision - % of positives correctly classified• Recall - % positive predictions correct• F1 – evenly weighted Precision and Recall• ROC – Left side is Threshold=1, Right side is Threshold=0• Recall Curve• AUC – area under curve across all thresholds, max = 1
• Precision/Recall – as threshold changes• Lift chart – “costed”
Demo: Public Web Service• Model Setup• Trained Model Block• Data Set Block• Score Block
• Adding Inputs / Outputs• Run• Deploy
Demo: Consume Web Service• Web page test• Excel • Code samples: C#, R, Python, etc.• REST
What we covered:• What’s Azure?• What’s Azure Machine Learning?• Getting Data• Model Building• Publishing as a Web Service• Consuming the Web Service
Conclusion• MS has thought through integration of analytics into systems• Input• Output
• New blocks added all the time• Re-calibration, re-deploy, versioning, etc. possible / automate-able• Powershell
• Metered/charged: storage, compute, database transaction units, bandwidth• Sell-able as a web service• Must be approved as a seller, have a pricing plan, approved as a service, etc.
Questions• Questions?
Resources• Azure Portal• portal.azure.com
• Selling a web service in the market• https://github.com/Azure/azure-content-nlnl/blob/master/articles/machine-l
earning/machine-learning-publish-web-service-to-azure-marketplace.md