siftr

If you work as an analyst, you probably shift projects often and need to get oriented in a new dataset quickly. siftr is an interactive tool that helps you find the column you need in a large dataframe using powerful 'fuzzy' searches.

It was designed with medical, census, and survey data in mind, where dataframes can reach hundreds of columns and millions of rows.

Installation

# CRAN soon

# Or install the live development version from Github.
remotes::install_github("DesiQuintans/siftr")

Starting siftr with every R session

For convenience, you can add siftr to your .Rprofile so that it is immediately available when you start R.

file.edit(file.path("~", ".Rprofile"))  # Opens your global .Rprofile for editing.

Add this line and save it:

options(defaultPackages = c('datasets', 'utils', 'grDevices', 'graphics', 'stats', 'methods', 'siftr'))

Functions in siftr

Function	Description
sift()	Search through a dataframe's columns.
sift.name()	Only search variable names (i.e. column names).
sift.desc()	Only search descriptive labels.
sift.factors()	Only search factor labels (and value labels).
save_dictionary()	Save the data dictionary for use with tsv2label
options_sift()	Get and set options related to how siftr functions.
mtcars_lab	A dataset bundled with the package for testing.

Ways of searching in siftr

1. Exact matching with or without regular expressions
2. Fuzzy matching with or without regular expressions
3. Orderless exact matching with or without regular expressions

Examples

library(siftr)
data(starwars, package = "dplyr")

By default, sift() searches for exact matches in a column's names, labels, levels, and unique values. As a convenience, you can type bare names in (i.e. color instead of "color") for simple queries.

sift(starwars, color)

#> ℹ Building dictionary for 'starwars'. This only happens when it changes.
#> ✔ Dictionary was built in 0.01 secs.
#> 
#> 4 hair_color
#>         Type: character         Missing: 5 %            All same? No            
#>         Peek: auburn, grey, grey, brown, blond, white, auburn, white, …
#> 5 skin_color
#>         Type: character         Missing: 0 %            All same? No            
#>         Peek: white, blue, grey, red, green-tan, brown, fair, blue, ye…
#> 6 eye_color
#>         Type: character         Missing: 0 %            All same? No            
#>         Peek: blue-gray, yellow, unknown, red, blue, gold, black, haze…
#> 
#> ✔ There were 3 results for query `color`.

As you can see, sift() returns lots of useful information about the variables it has found: The column number and name, its type, how much of it is NA/NaN, whether all of its values are the same, and a random peek at some of the column's unique values.

The .dist argument opts-in to approximate searching. It can take an integer (the number of characters that can be flexibly matched) or a double between 0 and 1 (e.g. 0.25 = 25% of the query pattern's length can be flexibly matched).

sift(starwars, homewolrd, .dist = 0.25)

#> 10 homeworld
#>         Type: character         Missing: 11 %           All same? No            
#>         Peek: Serenno, Trandosha, Aleen Minor, Cerea, Cato Neimoidia, …
#> 
#> ✔ There was 1 result for query `homewolrd`.

You can search with regular expressions, but these must be given as Character strings.

sift(starwars, "gr(a|e)y")

#> 4 hair_color
#>         Type: character         Missing: 5 %            All same? No            
#>         Peek: auburn, grey, grey, brown, blond, white, auburn, white, …
#> 5 skin_color
#>         Type: character         Missing: 0 %            All same? No            
#>         Peek: white, blue, grey, red, green-tan, brown, fair, blue, ye…
#> 6 eye_color
#>         Type: character         Missing: 0 %            All same? No            
#>         Peek: blue-gray, yellow, unknown, red, blue, gold, black, haze…
#> 
#> ✔ There were 3 results for query `gr(a|e)y`.

If you give multiple queries, then you will get an orderless look-around search.

sift(mtcars_lab, gallon, mileage)

#> ℹ Building dictionary for 'mtcars_lab'. This only happens when it changes.
#> ✔ Dictionary was built in 0.01 secs.
#> 
#> 2 mpg
#>     Mileage (miles per gallon)
#>         Type: double      Missing: 0 %      All same? No                        
#>         Peek: 15.2, 21.5, 15, 30.4, 16.4, 14.3, 24.4, 15.5, 19.2, 22.8…
#> 
#> ✔ There was 1 result for query `(?=.*gallon)(?=.*mileage)`.

Finally (and most powerfully), you can combine regular expressions and orderless look-around searches.

sift(starwars, color, "[a-z]{4}_")

#> 4 hair_color
#>         Type: character          Missing: 5 %            All same? No             
#>         Peek: blond, unknown, none, auburn, grey, blonde, brown, auburn,…
#> 
#> 5 skin_color
#>         Type: character          Missing: 0 %            All same? No             
#>         Peek: white, brown mottle, white, blue, fair, green, yellow, blu…
#> 
#> ✔ There were 2 results for query `(?=.*color)(?=.*[a-z]4_)`.

siftr works best on labelled data

sift() searches through these fields:

1. A column's name (colnames(df))
2. Its label (attr(col, "label"); placed by many packages including haven and labelled)
3. Its value labels (attr(col, "labels"); often hold-overs from SPSS or SAS datasets)
4. Its factor levels (levels(col))
5. Its unique values (unique(col)), sampled at random for large datasets

The more of these fields you can fill out, the more informative and powerful sift() will be.

siftr pairs well with one of my other packages, tsv2label, which can label, rename, and factorise a dataset using a plain text dictionary.