R Wrangling String Processing III

Separate with Regex

• The extract() function behaves similarly to the separate() function but allows extraction of groups from regular expressions.

Code:

# first example - normally formatted heights
s <- c("5'10", "6'1")
tab <- data.frame(x = s)

# the separate and extract functions behave similarly
tab %>% separate(x, c("feet", "inches"), sep = "'")
tab %>% extract(x, c("feet", "inches"), regex = "(\\d)'(\\d{1,2})")

# second example - some heights with unusual formats
s <- c("5'10", "6'1\"","5'8inches")
tab <- data.frame(x = s)

# separate fails because it leaves in extra characters, but extract keeps only the digits because of regex groups
tab %>% separate(x, c("feet","inches"), sep = "'", fill = "right")
tab %>% extract(x, c("feet", "inches"), regex = "(\\d)'(\\d{1,2})")

Code: Height Case study Composite

pattern <- "^([4-7])\\s*'\\s*(\\d+\\.?\\d*)$"

smallest <- 50
tallest <- 84
new_heights <- reported_heights %>% 
  mutate(original = height, 
         height = words_to_numbers(height) %>% convert_format()) %>%
  extract(height, c("feet", "inches"), regex = pattern, remove = FALSE) %>% 
  mutate_at(c("height", "feet", "inches"), as.numeric) %>%
  mutate(guess = 12*feet + inches) %>%
  mutate(height = case_when(
    !is.na(height) & between(height, smallest, tallest) ~ height, #inches 
    !is.na(height) & between(height/2.54, smallest, tallest) ~ height/2.54, #centimeters
    !is.na(height) & between(height*100/2.54, smallest, tallest) ~ height*100/2.54, #meters
    !is.na(guess) & inches < 12 & between(guess, smallest, tallest) ~ guess, #feet'inches
    TRUE ~ as.numeric(NA))) %>%
  select(-guess)

# Shortest Students
new_heights %>% arrange(height) %>% head(n=7)

String Splitting

• The function str_split() splits a string into a character vector on a delimiter (such as a comma, space or underscore). By default, str_split() generates a list with one element for each original string. Use the function argument simplify=TRUE to have str_split() return a matrix instead.
• The map() function from the purrr package applies the same function to each element of a list. To extract the ith entry of each element x, use map(x, i).
• map() always returns a list. Use map_chr() to return a character vector and map_int() to return an integer.

Code: String Splits

# read raw murders data line by line
filename <- system.file("extdata/murders.csv", package = "dslabs")
lines <- readLines(filename)
lines %>% head()

# split at commas with str_split function, remove row of column names
x <- str_split(lines, ",") 
x %>% head()
col_names <- x[[1]]
x <- x[-1]

# extract first element of each list entry
library(purrr)
map(x, function(y) y[1]) %>% head()
map(x, 1) %>% head()

# extract columns 1-5 as characters, then convert to proper format - NOTE: DIFFERENT FROM VIDEO
dat <- data.frame(parse_guess(map_chr(x, 1)),
                  parse_guess(map_chr(x, 2)),
                  parse_guess(map_chr(x, 3)),
                  parse_guess(map_chr(x, 4)),
                  parse_guess(map_chr(x, 5))) %>%
  setNames(col_names)
  
dat %>% head
  
# more efficient code for the same thing
dat <- x %>%
  transpose() %>%
  map( ~ parse_guess(unlist(.))) %>%
  setNames(col_names) %>% 
  as.data.frame() 

# the simplify argument makes str_split return a matrix instead of a list
x <- str_split(lines, ",", simplify = TRUE) 
col_names <- x[1,]
x <- x[-1,]
x %>% as_data_frame() %>%
  setNames(col_names) %>%
  mutate_all(parse_guess)

Recoding

• Change long factor names with the recode() function from the tidyverse.
• Other similar functions include recode_factor() and fct_recoder() in the forcats package in the tidyverse. The same result could be obtained using the case_when() function, but recode() is more efficient to write.

Code: Recoding

# life expectancy time series for Caribbean countries
library(dslabs)
data("gapminder")
gapminder %>% 
  filter(region=="Caribbean") %>%
  ggplot(aes(year, life_expectancy, color = country)) +
  geom_line()

# display long country names
gapminder %>% 
  filter(region=="Caribbean") %>%
  filter(str_length(country) >= 12) %>%
  distinct(country) 

# recode long country names and remake plot
gapminder %>% filter(region=="Caribbean") %>%
  mutate(country = recode(country, 
                          'Antigua and Barbuda'="Barbuda",
                          'Dominican Republic' = "DR",
                          'St. Vincent and the Grenadines' = "St. Vincent",
                          'Trinidad and Tobago' = "Trinidad")) %>%
  ggplot(aes(year, life_expectancy, color = country)) +
  geom_line()