Case Studies in advanced analytics with R
1 like1,026 views
A talk I gave at SQLDay 2017: About 1,5 years ago Microsoft finalised acquisition of Revolution Analytics – a provider of software and services for R. In my opinion this was one of the most important event for R community. Now it is crucial to present its capabilities to SQL Server community. It will be beneficial for both parties. I will present three case studies: cash optimisation in Deutsche Bank, midterm model for energy prices forecasting, workforce demand optimising. The case studies were implemented with our analytical workflow R Suite that will be also shortly presented.
![32
Availablewindpower[MWh]
Date](https://image.slidesharecdn.com/20170517sqldaywlogadvancedanalyticscasestudiesr-170604083247/85/Case-Studies-in-advanced-analytics-with-R-32-320.jpg)
![Introducing TiDB [Delivered: 09/25/18 at Portland Cloud Native Meetup]](https://cdn.slidesharecdn.com/ss_thumbnails/portlandk8smeetupintroducingtidb-180926052719-thumbnail.jpg?width=560&fit=bounds)
More Related Content
What's hot (20)
Similar to Case Studies in advanced analytics with R (20)
More from Wit Jakuczun (11)
Recently uploaded (20)
![Glary Utilities Pro 5.157.0.183 Crack + Key Download [Latest]](https://cdn.slidesharecdn.com/ss_thumbnails/artificialintelligence17-250529071922-ef6fe98e-thumbnail.jpg?width=560&fit=bounds)
Case Studies in advanced analytics with R
- 1. Case Studies in advanced analytics with R Wit Jakuczun, WLOG Solutions 2017-05-17
- 2. 2
- 3. 3 ~1995 R project has been started
- 5. 5 2006 I delivered my first project using R (~100 models for gas demand forecasting)
- 6. 6 2007 Revolution Analytics was founded.
- 7. 7 2015 Microsoft acquired Revolution Analytics.
- 8. 8 2017 R is (still) one of the most popular language for data science.
- 10. 10 What is so special about R? Low barriers to start using it Huge base of community Packages (CRAN/Github) Easy to build a team knowing R
- 11. 11 problem <- model %>% mzn_problem() %>% mzn_data_file(system.file("extdata", "examples", "sbprgeost.dzn", package = "minizincR")) %>% mzn_out_expr(var_name = "l") %>% mzn_out_expr(var_name = "roff", idxs = c("ROFF", "1..2")) %>% mzn_out_expr(var_name = "kind", idxs = c("BLOCK")) %>% mzn_out_expr(var_name = "coord", idxs = c("BLOCK", "1..2")) solver <- fzn_gecode(args = NULL) solution <- solver %>% solve(problem) print(solution) gg <- local({ k <- function(y) f(y) f <- function(x) if(x) x*k(x-1) else 1 }) gg(3)
- 12. 12 However, out-of-the-box R does not provide many essential features required in large-scale production deployment!
- 13. 13 “If you are using R and you think you’re in hell, this is a map for you.” Patrick Burns, “R Inferno”, 2011
- 14. WLOG R Suite™ How do we deliver R based solutions? 14
- 15. 15 CRAN (MRAN) Github Other R workspace Installed packages Local CRAN Source code repo
- 16. 16 CRAN (MRAN) Github Other R workspace Installed packages Local CRAN Source code repo
- 17. 17 Project A Project B Project C Production server ● Project independence ● Configuration and version management ● Version control ● Continuous integration&deployment
- 18. 18 WLOG R Suite™ Source code versioning Configuration management Continuous integration & deployment Seasoned R packages
- 19. Case Studies in advanced analytics with R 19
- 20. Cash optimisation in Deutsche Bank 20
- 21. 21 How much, how and when should I transfer between branches and vaults? Minimizing costs subject to (complex) business rules
- 22. 22
- 23. 23
- 24. 24 Historical data Cash flow “forecasting” Optimisation phase I Optimisation phase II Business rules Human intervention Results FONG
- 25. 25 Summary ● Our first large R project ● Around 20K LOC ● Webservice in C# ● R core is very stable ● Logic in packages ● No print allowed ● Project folder structure ● Configuration management ● (Almost) No REPL
- 26. Midterm model for energy prices forecasting 26
- 27. 27 What can be the energy price next year? Alternative (to the forward market) spot price forecast.
- 28. 28 Yesterday Today M+1-36, Y+1(2,3) Historical prices Wind speed Temperature Power availability Weather forecast? Weather forecast? Power availability
- 29. 29 Dane o rynku
- 30. 30 Date PricePLN/MWh ● (multi-)Seasonal ● Calendar effects ● Highly variable ● Spikes
- 33. 33
- 34. 34 Demand model Wind generation model Fundamental (economic) model Statistical layer
- 35. 35 Weather scenario 1 Coordination plans (PSE) Midterm model ... Weather scenario 1 Forecast 1 Forecast 1 Forecast 1
- 36. 36 M5 Base 2016 forward contract price and SPOT price scenarios Scenario S1 low temp, light wind S2 low temp, ave wind S3 low temp, strong wind S4 ave temp, light wind S5 ave temp, ave wind S6 ave temp, strong wind S7 high temp, light S2 wind S8 high temp, ave wind S9 high temp, strong wind
- 37. 37 ● Around 5K LOC ○ All Energy Pack is around 30K LOC ● Excel based UI ● logging ● data.table ● RCurl ● doParallel ● knitr Summary
- 39. 39 How many people should be in the location to generate optimal margin? Shop Warehouse Call center
- 40. 40 Traffic forecaster (R) Efficiency curves model (R) Workforce Demand Optimizer (R+CBC) UI Analytical data mart (Oracle)
- 42. 42 ● Around 15K LOC ● Integrated via DB ● First version of WLOG R Suite™ ● logging ● data.table ● RCurl ● doParallel ● knitr Summary
- 43. 43 What is so special about R? Glue language Fast prototyping in large scale Easy to build a team knowing R
- 45. 45 We are strategic partner in R based large scale analytics. Since 2005 with large scale R deployments with up-to 20-30K LOC.
- 46. 46 Understand why we are best for being Your partner in R for large scale analytics.
- 47. 47 Hire us for an analytical project
- 48. 48 Understand our analytical stack