Rise of the Machine (Learning) Azure ML in BI, in Apps, and as a product in Azure Marketplace

Size: px
Start display at page:

Download "Rise of the Machine (Learning) Azure ML in BI, in Apps, and as a product in Azure Marketplace"

Transcription

1 Rise of the Machine (Learning) Azure ML in BI, in Apps, and as a product in Azure Marketplace Greg R Beaumont November 18, linkedin.com/in/gregbeaumont

2 PASS Business Analytics Virtual Chapter Greg R linkedin.com/in/gregbeaumont

3 What Do I Hope You Get Out of this Session? How can predictive analytics and Azure ML add value to analytics? What is Azure ML, and why is it different from traditional machine learning tools? What are some examples of use cases for Azure ML? How do I get started with Azure ML and machine learning? Q & A

4 About Me I have worked on Microsoft BI stack projects since 2007 As GNet Group s North American Data Science Practice Lead I am fortunate to have worked with Azure ML since 2014 while it was still in preview Education: St. Mary s University of Minnesota B.A. in Biology/Pre-Med Carlson School of Management (University of Minnesota) MBA Self-Directed Learning Business Intelligence & Data linkedin.com/in/gregbeaumont

5 What is Machine Learning? Machine learning is a scientific discipline that explores the construction and study of algorithms that can learn from data. Such algorithms operate by building a model based on inputs and using that to make predictions or decisions, rather than following only explicitly programmed instructions.

6 Machine Learning, Data Mining, Data Science, AI, Big Data? How do they all Fit Together? Statistics Data Science Artificial Intelligence the study and design of intelligent agents and the creation of machines that think Predictive Analytics Machine Learning Big Data Massive volumes of Structured and also Unstructured Data Data Mining

7 Immanuel Kant s Insight into Reality applied to Machine Learning If our brain is a cup, we can only perceive the world in the way that it fills the cup Machine Learning algorithms perceive insights based upon the way that your data fills those algorithms

8 Existing Machine Learning Success Stories Speech recognition technology SPAM filters Recommendation engines Handwriting recognition News story clustering Credit fraud detection Self driving cars Robotics Image Recognition

9 Image Recognition: Goldendoodle dog

10 Image Recognition: Goldendoodle with a Schnauzer Cut

11 Decision Making Process Today Look at Data to Understand Business Issue What Information do I Know? Lessons Learned (Training the Business) Intuition Ask Others Wish I Could Know More Talk to Analysts Decision Ask IT for More Reports

12 Decision Making Process with Machine Learning (ML) Look at Data to Understand Business Issue What Information do I Know? Intuition Lessons Learned (Training the Business) Validate Intuition with ML Ask Others Wish I Could Know More Compare Outcomes to ML Predictions Decision Talk to Analysts Ask IT for More Reports

13 How Do We Find the Needle in the Haystack?

14 How Does a Frog Recognize Food? Intuition took millions of years to evolve. Business strategy evolves too fast for intuition to keep up! Businesses don t have time for the Evolutionary Learning of intuition. Businesses need Revolutionary Learning!

15 What is Machine Learning? (The Microsoft Perspective) Computing systems that become smarter with experience Computing systems that become smarter with experience Experience = past data + human input Delivering on one of the old dreams of Microsoft co-founder Bill Gates: Computers that can see, hear and understand. John Platt Distinguished scientist at Microsoft Research

16 Microsoft & Machine Learning Answering questions with experience Microsoft Research formed Hotmail launches Bing maps launches Bing search launches Kinect launches Skype Translator launches Azure Machine Learning launches Which is junk? What s the best way home? Which searches are most relevant? What does that motion mean? What is that person saying? What will happen next? John Platt, Distinguished scientist at Microsoft Research Machine learning is pervasive throughout Microsoft products.

17 Current State of the Business No improvement in generations Expensive Huge set-up costs of tools, expertise, and compute/storage capacity create unnecessary barriers to entry Siloed data Disconnected tools Deployment complexity Siloed and cumbersome data management restricts access to data Complex and fragmented tools limit participation in exploring data and building models Many models never achieve business value due to difficulties with deploying to production The Cloud Changes the Landscape

18 Decision Making Process with Machine Learning (ML) Look at Data to Understand Data Warehouse Business Issue What Information do I Know? Intuition Lessons Learned (Training the Business) Azure ML Model Validate Intuition with ML Build Model Train Model Test Model Tweak Model Score Model Ask Others Wish I Could Know More Compare Outcomes to ML Predictions Decision Reports Talk to Analysts Ask IT for More Reports Web UI

19

20 High Level Architecture Example Azure ML Model CRM Custom Reporting Leads / Inputs to Campaign Management or CRM ERP Data Warehouse / Data Mart Data Preparation Reports Data Warehouse Web Analytics SQL Server Excel Ad-hoc Reporting Web UI Azure HDInsight Data Sources

21 Smart buildings The Center for Building Performance and Diagnostics uses weather forecasts, real-time temperature reads, and behavioral research data to optimize building heating and cooling systems in real-time. User friendly set up and integration with existing systems Seamless data handling Accessible and easy to use across backgrounds Quickly compare algorithms The ease of implementation makes machine learning accessible to a larger number of investigators with various backgrounds even non-data scientists. Bertrand Lasternas Carnegie Mellon

22 Demand forecasting Pier 1 partnered with MAX451 to delight loyalty customers by using historical and behavioral data to predict what products they want next. Ease of use across skillsets Fast time to meaningful results Accessible via the cloud We are especially pleased that our analysts can focus on the results and not worry about the complex algorithms behind the scenes. Andrew Laudato Pier 1 Imports

23 Additional Use Case Scenarios Manufacturing Warranty & Claims Based upon features of past claims, predict future defect rates and trends Lean - Based upon results of samples taken during production, predict waste and batch quality Machine Component Data Monitor data from machine components to proactively monitor servicing and performance Healthcare Marketing Logistics Retail Using features of patient visit data, build a model that predicts high re-admission risk Use Azure ML to determine the mix of different diagnoses, lab data results, etc. to predict defined risks (e.g. for surgical complications, disease management, etc.) Track data from social media such as a Twitter API Determine mix of products that will maximize sales based upon weather forecasts Determine the effects of weather for just-in-time delivery, shipping on-time metrics, etc. Market Basket Analysis to determine mixes of products at different locations that combine to increase overall sales Stock & Inventory forecasting based upon weather trends and forecasts Attribution Analysis to determine best mix of web page activities to predict a sale

24 Common Types of Machine Learning Algorithms Ready to Go in Azure ML

25 Demo #1: Heat Stroke and Weather Data Set #1: HCUP SEDD (ER) Data for the State of Arizona Visit Level Data ICD9 codes, Locations, Visit level information Demographics such as Age, Gender, Race, Ethnicity Data Set #2: Weather Data from NOAA Daily temperature, precipitation, snowfall, etc by weather station Weather stations mapped to zip codes What factors impact expected heat stroke patients to the ER? Can we forecast the volume of heat stroke patients based upon weather forecasts?

26 Demo #2: Retail Marketing Campaign Data Set #1: Historical Customer Data Demographic Data Customers who Responded to a Previous Marketing Campaign are flagged Data Set #2: List of Potential Customers 18,000 Potential Customers within 10 Miles of a location queried from a large database Demographic Data $100,000 left in Sales & Marketing Budget $60 per Person Activity cost $90,000 in Total Net Profit needed in 4 weeks to meet sales goal Sales Team only large enough to reach 2,000 potential customers

27 Marketing Campaign Options by the Numbers ACTION 18,000 Total Sales Calls * $60 RESULT $1.1 Million Spend STATUS $100,000 Budget 1667 Sales Calls (Max Allowed by Budget) 47% are Buyers & $118k in Net Sales $118k Net Sales - $100k cost = $18k net Use Azure ML and Excel to Isolate 1,555 Likely Buyers 82% are Buyers & $191k in Net Sales $191k Net Sales - $93k cost = $97.6k net profit

28 Demo #3: Diagnostic Breast Cancer Wisconsin Data Data Set: 500+ biopsy results from breast tissue masses 10 different measurements for cell biopsies Actual data made public via UCI Machine Learning Repository Bache, K. & Lichman, M. (2013). UCI Machine Learning Repository [ Irvine, CA: University of California, School of Information and Computer Science Azure ML Model: Based upon cell biopsy measurements, are the cells benign or malignant? Uses for Azure ML Model: Analysis of historical pathology results Predict Benign or Malignant for new biopsy results with Azure ML Model published as a service A tool to assist and improve the pathology and diagnostic process

29 Azure ML Diagnostic Case Results Azure ML has robust algorithms that reliably predict outcomes Only 569 rows to train and test the Model 93% accuracy in predicting malignancy correctly Accuracy can improve over time Demo model was built in a few days by one resource Azure ML Model created without writing any code More Data, improved algorithms, new features Improve Outcomes Speed to Test and Validate Intuition Reduce Errors Eliminate Guesswork

30 How does Azure ML bring Machine Learning to the Masses? Feature Traditional Machine Learning Azure ML Strategic Advantage Price $$$$$ $ Affordable for the Masses Statistics Skill Set Programming Languages Infrastructure API to call the Model High Level of Statistics Required Usually require specialized language skills such as R and Python Usually Requires On Premise Hardware, Software, etc. Varies by product Get started with moderate understanding of Statistics using pre-packaged algorithms Models can be built without writing any code, but is also compatible with R and Python for advanced users Less ramp up required to start using the product to create value Less ramp up required to start using the product to create value 100% in the Azure cloud No hardware maintenance, no up front costs, no software upgrade efforts Call your Models using an API for single predictions or batch execution Call Azure ML from a web page, Excel, ETL package, mobile app, SSRS, third party tools

31 Traditional Utilization of Machine Learning Large High Tech Companies Large Low Tech Companies Mid Sized Companies Small Companies & Entrepreneurs Experienced ML Users New and Inexperienced ML Users

32 Azure ML Lowers the Barriers of Entry Large High Tech Companies Large Low Tech Companies Mid Sized Companies Small Companies & Entrepreneurs Experienced ML Users New and Inexperienced ML Users

33 Azure ML Marketplace Use pre-built Azure ML Models for your data with your solutions Market Basket Analysis (Frequently Bought Together) Recommendation Engine Anomaly Detection Text Analytics (Sentiment Analysis) Publish your own Azure ML Models to Azure Marketplace and license them out to others The App Model is now a reality for Machine Learning Models Companies can buy and sell Machine Learning Models Individual developers and entrepreneurs have a worldwide forum for their Models Machine Learning can now be Democratized

34 Azure ML Pricing Calculator

35 It s Time to View Issues Differently Just Add Azure ML! NOW Add Azure ML NOW Add Azure ML NOW Sales Force Dashboards Show Customer Sales to Goals Add Azure ML Use ML to determine which customers are most likely to buy and help you make goal Price Sale Items Based on Schedules and History Real Time Specials based on Inventory, Margins, Weather, etc. NOW Car Dealers Review Sales Data and KBB to Price Trade Ins Add Azure ML Estimate Used Car Sales Price and Time on Lot with ML Conduct Marketing Campaigns Based on Goals and Experience Plan Article Content, Hashtags, etc. using ML predictions

36 Q & A Greg R Beaumont Data Science Practice Lead GNet linkedin.com/in/gregbeaumont