Easing the Path to Production - APIs in R

Writing API's in R is a fast and efficient way to embed powerful analytic capabilities within your organisation

Transcript

1. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Easing the Path to Production Building API’s in R Mark Sellors – Mango Solutions
2. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Me • Head of Data Engineering • Devops/DataOps • Data Architecture • API’s • Analytic Maturity • Automation • Industrialisation
3. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Mango Solutions • R Specialists • Data Science Consulting • Data Science Training • Software Development Web: www.mango-solutions.com Twitter: @MangoTheCat
4. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com
5. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com
6. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com EARL
7. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Web: earlconf.com Twitter: @earlconf
8. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Agenda • API’s • Introducing plumber • How it works • Examples • Caveats • A simple case study • Summary
9. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Data Science + API’s are Changing the World
10. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Introducing plumber http://plumber.trestletech.com
11. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - simple function > adder <- function(a, b){ + a + b + } > adder(4, 3) [1] 7
12. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - add plumber #* @get /adder > adder <- function(a, b){ + as.numeric(a) + as.numeric(b) + }
13. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - running with plumber > library(plumber) > r <- plumb("my_api_file.R") > r$run(port=8000)
14. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - Accessing the API # curl http://localhost:8000/adder?a=4&b=3
15. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Demo 1 - Simple API
16. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - serving graphs
17. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Demo 2 - Tick API
18. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Why? What’s the point?
19. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Example - Shiny App
20. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Demo 3 - Currency Converter
21. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Moving beyond the shiny app • split the application up • turn the function into a general purpose service • refactor the shiny app to use the API
22. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Demo 4 - Currency API
23. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Caveats
24. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com When not to use plumber • Shiny is a better for many users/uses • It’s predominantly for API’s, not a full web stack solution
25. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Extra things to consider... • Security • input sanitisation • logging • monitoring • load balancing and state • blocking
26. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Insurance Claim Processing Case Study
27. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Claim processing is human powered, labour intensive and slow!
28. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com can help!
29. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Old workflow •Claim received •Passed to human processing team •Claim is processed •Result decided •Result conveyed to customer
30. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Easy acceptance Easy rejection The tricky middle area Human processing team Claims in How can we optimise?
31. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com New Workflow •Claim received •Claim form scanned •OCR performed and converted to XML representation •XML file is passed to an R service •Claim is compared against past claims and scored •Score determines next action: •Immediate payout •Immediate refusal •Passed to processing team for further action
32. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Workflow •Claim received •Claim form scanned •OCR performed and converted to XML representation •XML file is passed to an R service •Claim is compared against past claims and scored •Score determines next action: •Immediate payout •Immediate refusal •Passed to processing team for further action XML file is passed to an R service
33. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Any function you create in R can be turned into a service
34. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Corporate Network Website Inputs Results Inputs Results
35. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Creating an API in R •RServe •OpenCPU •Plumber •DeployR •Domino Data Lab •Others
36. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com > summary(talk) •Deploying R as a service (API) is increasingly straightforward •Creating an API provides a simple mechanism for exposing your functions to other users. •Plumber provides a really simple way to quickly create API’s using R •Put the business challenge before the analytic problem
37. Mark Sellors – Head of Data Engineering msellors@mango-solutions.com Thanks! @MangoTheCat @sellorm @earlconf