Pragmatic machine learning for the real world
•
5 j'aime•2,201 vues
A description of machine learning and prediction APIs followed by some real-world considerations on the deployment of predictive models and their integration in apps and businesses. These are illustrated with a churn prediction example. In this presentation you'll find some information about Microsoft Azure Machine Learning and how it compares to PredictionIO, an open source solution for creating prediction servers. I also gave an exclusive presentation of the Machine Learning Canvas, which you must fill in before any attempt at implementing a predictive system! Video coming soon...
Recommandé
Contenu connexe
En vedette
En vedette (20)
Similaire à Pragmatic machine learning for the real world
Similaire à Pragmatic machine learning for the real world (20)
Dernier
Dernier (20)
Pragmatic machine learning for the real world
- 4. –Waqar Hasan, Apigee Insights “Predictive is the ‘killer app’ for big data.”
- 5. –Mike Gualtieri, Principal Analyst at Forrester “Predictive apps are the next big thing in app development.”
- 7. Data
- 8. BUT
- 9. –McKinsey & Co. “A significant constraint on realizing value from big data will be a shortage of talent, particularly of people with deep expertise in statistics and machine learning.”
- 11. What the @#?~% is ML?
- 13. “How much is this house worth? — X $” -> Regression
- 14. Bedrooms Bathrooms Surface (foot²) Year built Type Price ($) 3 1 860 1950 house 565,000 3 1 1012 1951 house 2 1.5 968 1976 townhouse 447,000 4 1315 1950 house 648,000 3 2 1599 1964 house 3 2 987 1951 townhouse 790,000 1 1 530 2007 condo 122,000 4 2 1574 1964 house 835,000 4 2001 house 855,000 3 2.5 1472 2005 house 4 3.5 1714 2005 townhouse 2 2 1113 1999 condo 1 769 1999 condo 315,000
- 15. Bedrooms Bathrooms Surface (foot²) Year built Type Price ($) 3 1 860 1950 house 565,000 3 1 1012 1951 house 2 1.5 968 1976 townhouse 447,000 4 1315 1950 house 648,000 3 2 1599 1964 house 3 2 987 1951 townhouse 790,000 1 1 530 2007 condo 122,000 4 2 1574 1964 house 835,000 4 2001 house 855,000 3 2.5 1472 2005 house 4 3.5 1714 2005 townhouse 2 2 1113 1999 condo 1 769 1999 condo 315,000
- 17. Bedrooms Bathrooms Surface (foot²) Year built Type Price ($) 3 1 860 1950 house 565,000 3 1 1012 1951 house 2 1.5 968 1976 townhouse 447,000 4 1315 1950 house 648,000 3 2 1599 1964 house 3 2 987 1951 townhouse 790,000 1 1 530 2007 condo 122,000 4 2 1574 1964 house 835,000 4 2001 house 855,000 3 2.5 1472 2005 house 4 3.5 1714 2005 townhouse 2 2 1113 1999 condo 1 769 1999 condo 315,000
- 18. ML is a set of AI techniques where “intelligence” is built by referring to examples
- 21. “Which type of email is this? — Spam/Ham” -> Classification
- 22. WATCH OUT!
- 23. • Need examples of inputs AND outputs • Need enough examples
- 24. ??
- 25. Prediction APIs
- 27. HTML / CSS / JavaScript
- 28. HTML / CSS / JavaScript
- 29. squarespace.com
- 32. The two phases of machine learning: • TRAIN a model • PREDICT with a model
- 33. The two methods of prediction APIs: • TRAIN a model • PREDICT with a model
- 34. The two methods of prediction APIs: • model = create_model(dataset) • predicted_output = create_prediction(model, new_input)
- 35. from bigml.api import BigML # create a model api = BigML() source = api.create_source('training_data.csv') dataset = api.create_dataset(source) model = api.create_model(dataset) # make a prediction prediction = api.create_prediction(model, new_input) print "Predicted output value: ",prediction['object']['output'] http://bit.ly/bigml_wakari
- 38. Phrase problem as ML task Engineer features Prepare data (csv) Learn model Make predictions Deploy model & integrate pred Evaluate model Measure impact PREDICTIONAPIS
- 39. • Deployment to production? • Maintenance? • monitor performance • update with new data
- 43. • D: Data preparation • A: Algorithm • S: Serving • E: Evaluation
- 45. • Open source • Spark’s MLlib -> prediction server • Expose model as (scalable & robust) API • DASE framework
- 46. • Send new data/events to event server • Send prediction queries to engine
- 51. PredictionIO vs Azure ML on KDnuggets
- 53. • Who: SaaS company selling monthly subscription • Question asked: “is this customer going to leave within 1 month?” • Input: customer • Output: no-churn (negative) or churn (positive) • Data collection: history up until 1 month ago
- 54. Learning -> OK but How to represent customers? What to do after predicting churn?
- 55. Customer representation: • basic info (age, income, etc.) • usage of service (avg call duration, overcharges, leftover minutes/month, etc.) • interactions with customer support (how many, topics of questions, satisfaction ratings)
- 56. Taking action to prevent churn: • contact customer • switch to different plan • fix issues • give special offer
- 57. Measuring performance: • #TP, #FP, #FN • F-measure? • ROI • Compare to baseline
- 64. PREDICTIONS OBJECTIVES DATA BACKGROUND ENGINE SPECS INTEGRATION
- 65. PREDICTIONS OBJECTIVES DATA BACKGROUND End-user Value prop Sources ENGINE SPECS ML problem Perf eval Preparation INTEGRATION Using pred Learning model
- 66. PREDICTIONS OBJECTIVES DATA BACKGROUND 1 2 3 ENGINE SPECS 4 5 6 INTEGRATION
- 67. html
- 68. example: churn
- 69. End-user Value prop Sources -> events ML problem Perf eval Features Using pred Learning model DASE
- 70. Why fill in ML canvas? • target the right problem for your company • choose right algorithm, infrastructure, or ML solution • guide project management • improve team communication
- 72. Recap
- 73. • Create value from data with ML! • Creating and deploying models is easy(er)! • Good data is essential! • Use the ML canvas! • Go to PAPIs Connect!
- 74. Some real-world insights • Models that are easier to maintain cost less • Need to explain predictions? • One problem may call for another one…