DAS Bereichsmeeting: Introduction to R for Data Science

• Short overview: Regression courses on coursera
• Introduction to Data Science with R
  • Data handling with dplyr
  • Data visualisation with ggplot2
  • Model training and evaluation with caret
  • Presentation of results
• Getting started with R

• Regression models

  • focus on statistics
  • in R

• Advanced Linear Models for Data Science 1: Least Squares

  • focus on mathematics
  • in R

• Machine Learning: Regression

  • focus on algorithms
  • recommended to do in Python

• Base R can be extended by lots of great libraries (=packages)
  • CRAN: The Comprehensive R Archive Network
  • Install packages directly from github
• It provides the necessary tools for all important tasks required for data science
  • Import & tidy (=clean) data
  • Transform, visualise and model data
  • Communicate the results
• This presentation will focus on the highlighted tasks!

• There are tools for reading data from almost any source …
  • read.xlsx(), read.csv(), …
• Currently most useful is dplyr for accessing databases

conn <- RSQLServer::src_sqlserver("tauranga", database = "AdventureWorks2012")

emp_data <- conn %>% tbl("vEmployee")

dept_data <- conn %>% tbl("vEmployeeDepartment") %>%
  select(BusinessEntityID, Department, GroupName, StartDate)

dept_data %>% top_n(5) %>% collect()

# A tibble: 5 × 4
  BusinessEntityID Department                            GroupName
*            <int>      <chr>                                <chr>
1              234  Executive Executive General and Administration
2              286      Sales                  Sales and Marketing
3              288      Sales                  Sales and Marketing
4              285      Sales                  Sales and Marketing
5              284      Sales                  Sales and Marketing
# ... with 1 more variables: StartDate <chr>

jobs_per_region <- emp_data %>%
  left_join(dept_data, by="BusinessEntityID") %>%
  group_by(Department, CountryRegionName) %>% summarise(count=n()) %>% collect()

jobs_per_region

Source: local data frame [21 x 3]
Groups: Department [16]

                   Department CountryRegionName count
*                       <chr>             <chr> <int>
1                       Sales         Australia     1
2                       Sales            Canada     2
3                       Sales            France     1
4                       Sales           Germany     1
5                       Sales    United Kingdom     1
6            Document Control     United States     5
7                 Engineering     United States     6
8                   Executive     United States     2
9  Facilities and Maintenance     United States     7
10                    Finance     United States    10
# ... with 11 more rows

Recommended package: ggplot2 (grammar of graphics)

ggplot(house_prices, aes(x=YearBuilt, y=SalePrice, col=OverallQual)) + geom_point()

Many great extensions for ggplot2, e.g.: ggTimeSeries, gganimate. I frequently use ggpairs:

ggpairs(house_prices[,c("YearBuilt", "OverallQual", "SalePrice")])

caret package: Classification and Regression Training

Tools for

• data splitting
• preprocessing
• feature selection
• model tuning
• cross validation

233 model classes available for usage with the caret framework, e.g.:

• adaboost
• linear models, generalized linear models
• naive bayes
• partial least squares
• decision trees, random forests, …

cvSplits <- createFolds(house_prices$SalePrice, k = 10, returnTrain = TRUE )

lmFit <- train(SalePrice ~ YearBuilt + OverallQual, house_prices, method="lm", trControl=trainControl(index=cvSplits))
pred <- predict(lmFit)

ggplot(data.frame(y=house_prices$SalePrice, pred=pred), aes(x=pred, y=y)) + geom_point()

• rmarkdown reports
  • Hint: this presentation was done within RStudio!
  • \( \LaTeX\ \) can be used for equations
  • css can be used for styling
• R Notebooks
  • rmarkdown syntax
• shiny apps: interactive web applications
  • written in R
  • can be extended by: HTML, css, javascript, …
  • several options for hosting
  • gallery of shiny apps

• Loading data: many options

  • Loading data from databases: dplyr
  • Great flexibility w.r.t querying data.frames

• Visualising data: ggplot2

  • steep learning curve, but very powerful

• Training models: caret

• Reporting results: rmarkdown, R notebooks, shiny

• Garrett Grolemund, Hadley Wickham: R for Data Science
• A list of ggplot2 extensions
• tidyverse | tidyverse relase notes
• The caret package
• R Markdown Basics
• Very interactive R video lectures: datacamp.com
• RStudio cheatsheets

Thank you for your attention! Questions? Remarks?