Building a Machine Learning Recommendation Engine in SQL
3 likes2,230 views
This document discusses building machine learning recommendation engines using SQL. It begins with an overview of data and analytics trends including the convergence of operational and analytical databases. The rise of machine learning is then covered along with how databases are integrating machine learning capabilities. A live demo is presented using the Yelp dataset to build a recommendation engine directly in SQL, leveraging the database's extensibility, stored procedures, and user defined functions. The document argues that training can be done externally but operational scoring can and should be done directly in the database for real-time applications.
Building a Machine Learning Recommendation Engine in SQL
- 1. Building a Machine Learning Recommendation Engine in SQL @garyorenstein @memsql MemSQL 1
- 2. Today’s Talk 1. State of Data 2018 according to Gartner 2. Rise of Machine Learning 3. Live Demo - A SQL Recommendation Engine MemSQL 2
- 3. SECTION 1 The State of DataAccording to Gartner 2018 MemSQL 3
- 4. Hype Cycle for Data Management 26 July 2017 Donald Feinberg Adam M. Ronthal G00313950 MemSQL 4
- 5. MemSQL 5
- 6. Multimodel has the potential to support both relational and nonrelational use cases while reducing the number of disparate DBMS products in an organization. MemSQL 6
- 7. the idea of a Hadoop distribution will become obsolete before it reaches the Plateau of Productivity MemSQL 7
- 8. Penetration continues to increase and organizations should be evaluating these resources for — cost-efficiency — infrastructure simplification and — new use cases, such as Hybrid Transactional/ Analytical Processing (HTAP) MemSQL 8
- 9. Build Your Digital Business Platform Around Data and Analytics 31 January 2018 Andrew White W. Roy Schulte Roxane Edjlali Joao Tapadinhas Svetlana Sicular G00350435 MemSQL 9
- 10. Select Challenges Data and analytics investments that are tied to measurable business outcomes are more likely to produce reportable benefits. MemSQL 10
- 11. Magic Quadrant for Data Management Solutions for Analytics 13 February 2018 Adam M. Ronthal Roxane Edjlali Rick Greenwald G00326691 MemSQL 11
- 12. We define four primary use cases for DMSAs that reflect this diversity of data and use cases: — Traditional data warehouse — Real-time data warehouse — Context-independent data warehouse — Logical data warehouse MemSQL 12
- 13. MemSQL 13
- 14. MemSQL 14
- 15. Real-Time Data Warehouse This use case adds a real-time component to analytics use cases, with the aim of reducing latency — the time lag between when data is generated and when it can be analyzed. MemSQL 15
- 16. MemSQL 16
- 17. Other Vendors to Consider for Operational DBMSs 23 November 2017 Donald Feinberg Merv Adrian Nick Heudecker G00327284 MemSQL 17
- 18. Other Vendors to Consider for Operational DBMSs Actian Aerospike Alibaba Cloud Altibase ArangoDB Cloudera Clustrix Couchbase FairCom Fujitsu General Data Technology Hortonworks MariaDB MemSQL MongoDB Neo4j NuoDB Percona Redis Labs SequoiaDB TmaxSoft VoltDB MemSQL 18
- 19. Other Vendors to Consider for Operational DBMSs also listed as Challenger or Leader in the Magic Quadrant for Data Management Solutions for Analytics MemSQL MemSQL 19
- 20. MemSQL 20
- 21. Over the next five years, the OPDBMS and DMSA markets converge to a single DBMS market. MemSQL 21
- 22. Look to your operational DBMS vendor for both transactional and analytical workloads. MemSQL 22
- 23. SECTION 2 Rise of Machine Learning MemSQL 23
- 24. MemSQL 24
- 25. MemSQL 25
- 26. MemSQL 26
- 27. MemSQL 27
- 28. MemSQL 28
- 29. MemSQL 29
- 30. 2018 Outlook Survey MemSQL and O’Reilly 1600+ respondents memsql.com/MLsurvey MemSQL 30
- 31. MemSQL 31
- 32. MemSQL 32
- 34. MemSQL 34
- 35. MemSQL 35
- 36. MemSQL 36
- 37. MemSQL 37
- 38. MemSQL 38
- 39. MemSQL 39
- 40. MemSQL 40
- 41. MemSQL 41
- 42. MemSQL 42
- 43. MemSQL 43
- 44. MemSQL 44
- 45. MemSQL 45
- 46. MemSQL 46
- 47. MemSQL 47
- 48. SECTION 3 DEMO with Yelp Dataset MemSQL 48
- 49. MemSQL 49
- 50. MemSQL 50
- 51. MemSQL 51
- 52. MemSQL 52
- 53. Can you build a machine learning recommendation engine in SQL? Yes MemSQL 53
- 54. Can you build a machine learning recommendation engine in SQL? Yes Should you? For training? Maybe, maybe not. For Operational Scoring? Absolutely! MemSQL 54
- 55. MemSQL 55
- 56. MemSQL 56
- 57. Secret Weapons to Machine Learning in SQL — Extensibility — Stored Procedures — User Defined Functions — User Defined Aggregates — DOT_PRODUCT — Compare two vectors MemSQL 57
- 58. MemSQL 58
- 59. MemSQL 59
- 60. Sequel Pro Mac app for MySQL databases MemSQL 60
- 61. MemSQL in one slide — Distributed SQL database — Massively parallel, lock-free, fast — Full ACID features — In-memory and on-disk — JSON, key-value, geospatial, full-text search — Robust security — Built for transactions and analytics MemSQL 61
- 62. MemSQL 62
- 63. MemSQL 63
- 64. Why do ML in SQL? — Train in any number of systems — Score in the database for applications from real-time drilling to fraud detection to personalization — Complete certain functions within the database to radically simplify operational infrastructure MemSQL 64
- 65. “It is a fine line between a well executed SQL query on live data and ML/AI” MemSQL 65
- 66. MemSQL 66
- 67. Thank you! Please visit our booth www.memsql.com @garyorenstein @memsql MemSQL 67
- 68. Abstract: Building a Machine Learning Recommendation Engine in SQL Modern businesses constantly seek deeper customer relationships and more compelling experiences. To accomplish this, companies are looking to machine learning and artificial intelligence solutions; however, that often involves a host of new systems and approaches. With a modern database architecture, it is possible to build compelling machine learning solutions with SQL, deliver real-time engagements, and rapidly move to operational applications. See live, how a modern database can accomplish these feats within a single integrated solution. MemSQL 68