r-cheatsheet

Cheatsheet for R & Rstudio

?<anything-here> # Help/manual for that thing

Packages

Install packages once from CRAN/Bioconductor/devtools

# CRAN
install.packages("dplyr")

# Github via devtools (Install Devtools dependecies first)
install_github()

Load package each time we use RStudio

library(dplyr)

Data Exploration

Data Loading

Consider doing data manipulation in SAS/SQL/Python/MapReduce etc., as R is not great wth manipulations of large-scale data.

Text File Encodings

Unless it's in ASCII, check encoding first with cmd file or od.

Note: BOMs in UTF-8 files cause issues when moving between Windows vs. UNIX, so ensure you know whether it has it beforehand.

Loading Local Data

CSV

mydata <- read.csv("filename.txt")

More Generic

mydata <- read.table("filename.txt", header = TRUE, sep = "\t",
           na.strings = "NA", fileEncoding = "", encoding = "unknown")

Fixed-width-column formats

read.fwf

Matrices

A <- matrix(scan("matrix.dat", n = 200*2000), 200, 2000, byrow = TRUE)

Viewing Data

We are going to use the made up sample-data dataset.

# View the Data
sample-data

# First line
first(sample-data)

# Dimensions
dim(sample-data)

# Column Names
names(sample-data)

# Column Structure & Data Type
str(sample-data)

# Investigate the structure of specific columns
sample-data %>% 
  select(column-name, column-b-name) %>% 
  str()

# View a column
sample-data$column-name

# Take a Sample (dplyr)
sample_n(sample-data)

Summarising Data

View as a table:

table(object/list)

Sample Size:

• Sample Size - n()

Measures of Centre:

(Often has option to exclude null values by setting argument "na.rm = TRUE")

• Mean - mean()
• Median - median()
• Mode:

getmode <- function(v) {
   uniqv <- unique(v)
   uniqv[which.max(tabulate(match(v, uniqv)))]
}

result <- getmode(vector)
print(result)

Measures of Spread:

• Range - range()
• Minimum - min()
• Maximum - max()
• Standard Deviation - sd()
• Variance - var()
• IQR - IQR()

Create a list of summary metrics

# Dplyr function
sample-data %>%
summarise(mean_dd = mean(column-name), sd_dd = sd(column-name), n = n())

Tabulate Categorical Data:

sample-data %>% 
  group_by(column-name) %>% 
  summarise(count = n())

Selecting & Grouping Data

Select Top n

head(n)

Filter

# Dplyr function 
# AND
new-data-frame <- sample-data %>%
  filter(column-name-a == "text", column-name-b == 999)

# OR  
new-data-frame <- sample-data %>%
  filter(column-name-a == "text" | column-name-b == 999)

# Null values
new-data-frame <- sample-data %>%
  filter(!is.na(column-name))

Select

new-data-frame <- sample-data %>%
  select(column-name-a, column-name-b)

Group

sample-data %>%
  group_by(column-name-a) %>%
  summarise(mean = mean(column-name-b), sd = sd(column-name-b), n = n())

Visualising Data

We use the ggplot2 package for visualisation

Example Scatter Plot

ggplot(data = sample-data, aes(x = column-name-a, y = column-name-b)) +
  geom_point() 

• Scatter Plot - geom_point()
• Line Plot - geom_line()
• Histogram -geom_histogram() or stat_bin()
  • Args: # bins: bins, bin width: binwidth
• Box Plot - geom_boxplot()
  • Requires categorical variable on the x axis, so factor() if variable is not already factor type.
• Bar Chart - geom_bar()
  • Stacked Bar Chart: aes(fill = column-name-a)

Manipulating Data

Standard Operators:

• Assignment operator <-
• Piping operator %>%

Many of the data manipulation verbs in this section come from the dplyr package.

• arrange()
• distinct()
• mutate()

Adding Variables

Adds a new column called output-column-name to sample-data

sample-data <- sample-data %>%
  mutate(output-column-name = column-name-a + column-name-b)

Rearrange Data

Arrange a column in descending order

sample-data <- sample-data %>% arrange(desc(column-name))

Change data types

Change a factor type column into numeric data type (via character data type). (Often useful for ordinal categorical variables)

sample-data <- sample-data %>%
  mutate(output-column-name = as.numeric(as.character(column-name)))

Feature Engineering

• Ternary If/Else - ifelse(column-name < 1, "FOO", "BAR"))

Simulations

Sampling

coin_outcomes <- c("heads", "tails")  #c() - combine values into vector/list
sim_unfair_coin <- sample(coin_outcomes, size = 1000, replace = TRUE, prob = c(0.2, 0.8))