Agenda

• R
• R + Microsoft
• RefresheR
• Super R
• Develop R
• Administer R
• Wrap up

R

R is an integrated suite of software facilities for data manipulation, calculation and graphical display

• Open source
• In-memory & single-core (by default)
• Multi-platform
• Extensible environment
• Delivered by the R Foundation, supported by the R Consortium, grown by R developers
• r-project.org

Setup on Windows

• Get R
• Get R Tools
• Get RStudio or Visual Studio + VS R Tools

There’s a package for that

• R is extended with packages of functionality
• More than 8,000 active packages according to MetaCRAN
• Comprehensive R Archive Network (CRAN)
• Github via devtools

How it hangs together

How it hangs together

Next steps

1. Check out the R website
2. Install R & RStudio (or R Tools for Visual Studio)

Microsoft R Server

• Formerly, Revolution R for Enterprise
• Available on Windows and Linux
• Specialised connectors available for SQL Server, Hadoop, Oracle, & Teradata

AzureML

• Uses R throughout

PowerBI

• R as a data source
• Use R for data visualisations
• Any R variant

SQL Server

• Call R from inside SQL

Syntax

Objects

• Vector
• Matrix
• Array
• List
• data.frame

Next steps

• Go through the first sections of the R for Microsoft Data Platform pre-con
• Try the datacamp R intro course

data.table

data.table

1. Super-fast slicing & dicing
2. Low memory footprint vs data.frames
3. Fast joins
4. Auto-indexing
5. Many saved characters!
6. Active dev

data.table

The “Hadleyverse”

Hadley Wickham, insanely prolific developer of R packages has produced a great ecosystem:

• httr
• ggplot2
• purrr
• readxl
• haven
• dplyr
• rvest
• …

dplyr

• filter() (and slice())
• arrange()
• select() (and rename())
• distinct()
• mutate() (and transmute())
• summarise()
• sample_n() and sample_frac()

dplyr

1. Relatively clear verbs
2. Quite easy to get started with
3. Verbose

ggplot2

ggplot2

1. Clean conceptual implementation
2. Highly customisable
3. Simple to start with
4. Big ecosystem

ggplot2

caret

A single point of contact for myriad statistical & machine learning algorithms

• Plus handy data prep!
• Find out more at caret homepage

AzureML

Interact with AzureML in R

1. No GUI
2. All the development environments of R
3. Consume AzureML webservice

rmarkdown

Write markdown, interweave code

1. Data provenance
2. Ease of documenting & developing solutions
3. Extendible & customisable

miniCRAN

Make your own CRAN

1. R interface
2. Control over packages
3. Internal package deployment

Next steps

1. Play with the popular packages
2. Look at CRAN task views
3. Ask for recommendations!

Utility packages

• testthat
• devtools (useful for other things)
• microbenchmark
• packrat

Package development

library(devtools)
pkg<-"newPackage"
create(pkg)

library(devtools) # Open the project!
add_test_infrastructure() # Add unit test framework
add_travis() # Add CI framework
use_vignette() # Add folder for macro-level help files
use_package_doc() # Add file for providing info about your package
use_readme_rmd() # Make a README!

What to unit test

1. A single set of values that represent “normal” and expect this matches a correct answer
2. A dataset that represents “normal” and expecting this to match correct answers
3. Various permutations of bad input values and expect errors (ideally specific error messages)
4. Edge cases that cover extreme values or boundaries for any inequalities or conditions
5. Any bugs or compatibility issues

How to test

myfunc<-function(a=1,b=2,c="blah"){
stopifnot(is.numeric(a), is.numeric(b), is.character(c))
d<-ifelse(a<0,a*b,a/b)
e<-paste0(c,d)
return(e)
}
library(testthat)
# Add a high-level name for group of tests, typically the function name
context("myfunc")

# Simplest test
test_that("Defaults return expected result",{
  result<-myfunc()
  check<-"blah0.5"
  expect_equal(result,check)
})

Next steps

1. Read Hadley’s Package Development guide
2. Practice!

Infrastructure

• Windows or Linux?
• Linux!
• Because …
• RStudio Server
• System Dependencies

Infrastructure

• High RAM
• CPU should be decent
• Physical storage can be low
• Virus scanning not a problem
• AD auth (fairly simple)
• File share availability
• Backups or scripted builds?

R

• Microsoft R or base R?
• It depends!
• Need cutting edge R & packages?
• Need extra computational speed or connectors?

R & Packages

• Regular updates required
• Unit tests of key behaviours should be done
• Dev environments always a plus!
• miniCRAN

Deployment

• Ad-hoc facilities
• Via Continuous Deployment

Next steps

• Learn some Linux fundamentals
• Determine what sort of capacity requirements are needed

Covered

• R overview
• R + Microsoft
• RefresheR
• Super R
• Develop R
• Administer R

Next steps

• Get more detail from my training materials
• Follow up the Next Steps sections
• Let me know how you get on: @SteffLocke

Q&A