cleaner

Fast and Easy Data Cleaning


License
GPL-2.0

Documentation

% cleaner

cleaner: Fast and Easy Data Cleaning

Website of this package: https://msberends.github.io/cleaner

CRAN_Badge

The small R package for cleaning and checking data columns in a fast and easy way. Relying on very few dependencies, it provides smart guessing, but with user options to override anything if needed.

It also provides two new data types that are not available in base R: currency and percentage.


Contents:


Why this package

As a data scientist, I'm often served with data that is not clean, not tidy and consequently not ready for analysis at all. For tidying data, there's of course the tidyverse (https://www.tidyverse.org), which lets you manipulate data in any way you can think of. But for cleaning, I think our community was still lacking a neat solution that makes data cleaning fast and easy with functions that kind of 'think on their own' to do that.

If the CRAN button at the top of this page is green, install the package with:

install.packages("cleaner")

Otherwise, or if you are looking for the latest stable development version, install the package with:

install.packages("remotes") # if you haven't already
remotes::install_github("msberends/cleaner")

How it works

This package provides two types of functions: cleaning and checking.

Cleaning

Use clean() to clean data. It guesses what kind of data class would best fit your input data. It calls any of the following functions, that can also be used independently. They always return the class from the function name (e.g. clean_Date() always returns class Date).

  • clean_logical() for values TRUE/FALSE. You only define what should be TRUE or FALSE and it handles the rest for you. At default, it supports "Yes" and "No" in the following languages: Arabic, Bengali, Chinese (Mandarin), Dutch, English, French, German, Hindi, Indonesian, Japanese, Malay, Portuguese, Russian, Spanish, Telugu, Turkish and Urdu. This covers at least two-third of the world population (Ulrich Ammon et al., University of Düsseldorf).

    # English
    clean_logical(c("Yes", "No", "Invalid", "Unknown"))
    #> [1]  TRUE FALSE    NA    NA
    
    # French
    clean_logical(c("Oui, c'est ca", "Non, pas encore")) 
    #> [1]  TRUE FALSE
      
    # Indonesian
    clean_logical(c("ya :)", "tidak :("))
    #> [1]  TRUE FALSE

    If you define the true and false parameters yourself, they will be interpreted as regular expressions:

    clean_logical(x = c("Positive", "Negative", "Unknown", "Unknown"),
                  true = "pos",
                  false = "neg")
    #> [1]  TRUE FALSE    NA    NA
    
    clean_logical(x = c("Probable", "Not probable"),
                  true = ".*",
                  false = "not")
    #> [1]  TRUE FALSE
  • clean_factor() for setting and redefining a factor. You can use regular expressions to match values in your data to set new factor levels.

    gender_age <- c("male 0-50", "male 50+", "female 0-50", "female 50+")
    gender_age
    #> [1] "male 0-50"   "male 50+"    "female 0-50" "female 50+"
    
    clean_factor(gender_age, levels = c("M", "F"))
    #> [1] M M F F
    #> Levels: M F
    
    clean_factor(gender_age, levels = c("Male", "Female"))
    #> [1] Male   Male   Female Female
    #> Levels: Male Female
    
    clean_factor(gender_age, levels = c("0-50", "50+"), ordered = TRUE)
    #> [1] 0-50 50+  0-50 50+ 
    #> Levels: 0-50 < 50+

    You can also name your levels to let them match your values. They support regular expressions too:

    clean_factor(gender_age, levels = c("female" = "Group A", 
                                        "male 50+" = "Group B",
                                        ".*" = "Other"))
    #> [1] Other   Group B Group A Group A
    #> Levels: Group A Group B Other
  • clean_Date() for any type of dates. This could be dates imported from Excel, or any combination of days, months and years. For convenience, the format parameter understands the date format language of Excel (like d-mmm-yyyy) and transforms it internally to the human-unreadable POSIX standard that R understands (%e-%b-%Y):

    clean_Date("13jul18", "ddmmmyy")
    #> [1] "2018-07-13"
    
    clean_Date("12-06-2012")
    #> (assuming format 'dd-mm-yyyy')
    #> [1] "2012-06-12"
    
    clean_Date("14 August 2010")
    #> (assuming format 'dd mmmm yyyy')
    #> [1] "2010-08-14"
    
    clean_Date(38071)
    #> (assuming Excel format)
    #> [1] "2004-03-25"

    The function to transform d-mmm-yyyy to %e-%b-%Y is available as format_datetime() to users. This makes it possible to use it in other date functions too:

    as.Date("12-13-14", format = format_datetime("mm-yy-dd"))
    #> [1] "2013-12-14"
  • clean_POSIXct() to remove all non-date/time characters and transform to a date/time element. It automatically adds the systems timezone, which can be changed by the user:

    a <- clean_POSIXct("Created log on 2019/04/11 11:23 by user Joe")
    a
    #> "2019-04-11 11:23:00 CEST"
    
    b <- clean_POSIXct("Log am 2019.04.11 11:23 erstellt", tz = "US/Michigan")
    b
    #> "2019-04-11 11:23:00 EDT"
    
    difftime(a, b)
    #> Time difference of -6 hours
  • clean_numeric() to remove all non-numbers from cluttered input text. It understands usage of dots and comma's in different languages:

    clean_numeric(c("$ 12,345.67",
                    "€ 12.345,67",
                    "12,345.67",
                    "12345,67"))
    #> [1] 12345.67 12345.67 12345.67 12345.67
    
    clean_numeric("qwerty123456")
    #> [1] 123456
    
    clean_numeric("Positive (0.143)")
    #> [1] 0.143
  • clean_character() to remove all obvious non-characters from cluttered input text:

    clean_character("qwerty123456")
    #> [1] "qwerty"
    
    clean_character("Positive (0.143)")
    #> [1] "Positive"

    You can define yourself what should be removed using the remove argument, with regular expressions:

    clean_character(x = c("Model: Pro A1          ",
                          "Model specified: Pro A1",
                          "       Pro A1          "), 
                    remove = "^.*:")
    #> [1] "Pro A1" "Pro A1" "Pro A1"
  • clean_percentage() to use the new percentage class that comes with this package. It prints numeric values as percentages using as.percentage():

    as.percentage(c(0.25, 2.5, 0.025))
    #> [1]  25.0% 250.0%   2.5%
    
    sum(as.percentage(c(0.25, 2.5, 0.025)))
    #> [1] 277.5%
    
    clean_percentage("PCT: 0.143")
    #> [1] 14.3%
  • clean_currency() to use the new currency class that comes with this package. It transforms the input with clean_numeric() first, after which it will be transformed with as.currency(), guessing the currency symbol based on your system locale:

    clean_currency(c("Jack sent £ 25", "Bill sent £ 31.40"))
    #> [1] `GBP 25.00` `GBP 31.40`
    
    received <- clean_currency(c("Received $25", "Received $31.40"))
    received
    #> [1] `USD 25.00` `USD 31.40`
    
    sum(received)
    #> [1] `USD 56.40`
    
    format(sum(received), 
           currency_symbol = "", decimal.mark = ",")
    #> [1] "EUR 56,40"

    This new class also comes with support for printing in tibbles, used by the tidyverse:

    library(dplyr)
    tibble(money = c("Jack sent £ 25", "Bill sent £ 31.40")) %>% 
      mutate(mutate_cleaner = clean_currency(money))
    #> # A tibble: 2 x 2
    #>   money               mutate_cleaner
    #>   <chr>                  <crncy/GBP>
    #> 1 Jack sent £ 25               25.00
    #> 2 Bill sent £ 31.40            31.40

Other cleaning

  • Use format_names() to quickly and easily change names of data.frame columns, lists or character vectors.

    df <- data.frame(old.name = "test1", value = "test2")
    format_names(df, snake_case = TRUE)
    format_names(df, camelCase = TRUE)
    format_names(df, c(old.name = "new_name", value = "measurement"))
    
    library(dplyr)
    starwars %>% 
      format_names(camelCase = TRUE) %>% # changes column names
      mutate(name = name %>% 
               format_names(snake_case = TRUE)) # changes values in column
  • Use the generic function na_replace() to replace NA values in any data type. Its default replacement value is dependent on the data type that is given as input: 0 for numeric values and class matrix, FALSE for class logical, today for class Date, and "" otherwise.

    na_replace(c(1, 2, NA, NA))
    #> [1] 1 2 0 0
    na_replace(c(1, 2, NA, NA), replacement = -1)
    #> [1]  1  2 -1 -1
    na_replace(c(1, 2, NA, NA), replacement = c(0, -1))
    #> [1]  1  2  0 -1
    
    na_replace(c("a", "b", NA, NA))
    #> [1] "a" "b" ""  ""

    It also supports replacing NAs in complete data sets and supports grouped variables used by the dplyr package:

    library(dplyr)
    starwars %>% 
      na_replace(hair_color) # only replace NAs in this column
      
    starwars %>% 
      na_replace() # replace NAs in all columns ("" for hair_color and 0 for birth_year)
      
    starwars %>%
      group_by(hair_color) %>%
      na_replace(hair_color, replacement = "TEST!") %>%
      summarise(n = n())
  • Use the function format_p_value() to format p values according to the international APA guideline. It tries to round to two decimals, but has a exception for values that would round to alpha (defaults to 0.05):

    format_p_value(c(0.345678, 0.123))
    #> [1] "0.35" "0.12"
    
    # a value of 0.0499 must not be "0.05", but is not "0.049" either,
    # so the function will add as many decimals as needed:
    format_p_value(0.04993) 
    #> [1] "0.0499"

Checking

The easiest and most comprehensive way to check the data of a column/variable is to create frequency tables. Use freq() to do this. It supports a lot of different classes (types of data), weights, and is even extendible by other packages. In markdown documents (like this README file), it formats as real markdown.

freq(unclean$gender)

Frequency table

Class: character
Length: 500
Available: 500 (100%, NA: 0 = 0%)
Unique: 5

Shortest: 1
Longest: 6

Item Count Percent Cum. Count Cum. Percent
1 male 240 48.0% 240 48.0%
2 female 220 44.0% 460 92.0%
3 man 22 4.4% 482 96.4%
4 m 15 3.0% 497 99.4%
5 F 3 0.6% 500 100.0%

Clean it and check again (using markdown = FALSE to show how it would look in the R console):

freq(clean_factor(unclean$gender, 
                  levels = c("^m" = "Male", "^f" = "Female")),
     markdown = FALSE)
#> Frequency table 
#> 
#> Class:      factor (numeric)
#> Length:     500
#> Levels:     2: Male, Female
#> Available:  500 (100%, NA: 0 = 0%)
#> Unique:     2
#> 
#>      Item      Count   Percent   Cum. Count   Cum. Percent
#> ---  -------  ------  --------  -----------  -------------
#> 1    Male        277     55.4%          277          55.4%
#> 2    Female      223     44.6%          500         100.0%

This could also have been done with dplyr syntax, since freq() supports tidy evaluation:

unclean %>% 
  freq(clean_factor(gender,
                    levels = c("^m" = "Male", "^f" = "Female")))
# or:
unclean %>% 
  pull(gender) %>% 
  clean_factor(c("^m" = "Male", "^f" = "Female")) %>% 
  freq()

Speed

The cleaning functions are tremendously fast, because they rely on R's own internal C++ libraries:

# Create a vector with 500,000 items
n <- 500000
values <- paste0(sample(c("yes", "no"), n, replace = TRUE), 
                 as.integer(runif(n, 0, 10000)))

# data looks like:
values[1:3]
#> [1] "no3697"  "yes1906" "yes6738"

clean_logical(values[1:3])
#> [1] FALSE  TRUE  TRUE

clean_character(values[1:3])
#> [1] "no"  "yes" "yes"

clean_numeric(values[1:3])
#> [1] 3697 1906 6738

# benchmark the cleaning based on 10 runs and show it in seconds:
microbenchmark::microbenchmark(logical = clean_logical(values),
                               character = clean_character(values),
                               numeric = clean_numeric(values),
                               times = 10,
                               unit = "s")
#> Unit: seconds
#> expr            min        lq      mean    median        uq       max neval
#> logical   0.2846163 0.2925479 0.3076008 0.3100244 0.3189712 0.3269428    10
#> character 0.4522698 0.4593437 0.4734631 0.4636837 0.4888959 0.5303473    10
#> numeric   0.6428362 0.6476207 0.6618845 0.6542312 0.6778215 0.6897005    10

Cleaning 500,000 values (!) only takes 0.3-0.6 seconds on our system.

Invalid regular expressions

If invalid regular expressions are used, the cleaning functions will not throw errors, but instead will show a warning and will interpret the expression as a fixed value:

clean_character("0123test 0123[a-b] ")
#> [1] "test ab"

clean_character("0123test 0123[a-b] ", remove = "[a-b]")
#> [1] "0123test 0123[-]"

clean_character("0123test0123", remove = "[a-b")
#> [1] "0123test 0123]"
#> Warning message:
#> invalid regular expression '[a-b', reason 'Missing ']'' - now interpreting as fixed value