2. Mission: what is the best way of getting
Value out of Data?
Data Insights or Products
3. The old way of doing things …
1. Obtain data and build a model/viz
Then:
2a. Write a Report
Or, rarely:
2b. Collaborate with software engineers
… to build UI, pipelines, databases
… to create an infrastructure for it
4. The new way:
2c: “Productivization toolkits”
Related solutions
Rook, Flask, Plotly/Dash, Tableau,
Spyre
10. R Shiny overview
● Shiny: an open-source R package by the people in RStudio
● Shinyapps.io: a cloud platform as a service
● Two main parts: server.R and ui.R
● input$foo = a variable containing input from the UI
● output$bar = a variable containing output data
● The app is event-driven by changes in these values
13. More on Shiny: Shiny deployment
1. Run on your desktop for testing
2. Run on your server ⟺ client devices
3. Run on shinyapps.io cloud ($$) ⟺ client devices
14. Leaflet library
● Javascript library for maps
● Well-integrated with shiny
● Draws on OpenStreetMaps
● Bindings in R and Python
18. Behind the scenes in VectorPoint
● AWS RDS/PostgreSQL database
● Shinyapps.io cloud platform
● Desktop-based
spatial predictive model
19. Successes
● Rapid development of an app (1 month to a prototype)
● Successful field tool used in Peru
● Easy maintenance
20. Implementation challenges with R/shiny
● R is not an easy language to learn
● Javascript is invisible until it breaks
● Code could become a messy mix of R & Javascript
21. ● Narrow focus: putting your R code online / on the cloud
● Client must be online
● Security is difficult
● Requires a UI
● R limits speed
Limitations of shiny
23. Conclusions
Shiny (and similar frameworks) make data scientists more valuable:
sasha.gutfraind@uptake.com
● Deliver much more than a model/report!
● Enables data collection, visualization, collaboration, prototyping
● Without the cost of traditional software engineering