Unpivot Complex and Irregular Data Layouts

Tools for converting data from complex or irregular layouts to a columnar structure. For example, tables with multilevel column or row headers, or spreadsheets. Header and data cells are selected by their contents and position, as well as formatting and comments where available, and are associated with one other by their proximity in given directions. Functions for data frames and HTML tables are provided.

unpivotr deals with non-tabular data, especially from spreadsheets. Use unpivotr when your source data has any of these ‘features’:

• Multi-headered hydra
• Meaningful formatting
• Headers anywhere but at the top of each column
• Non-text headers e.g. dates
• Other stuff around the table
• Several similar tables in one sheet
• Sentinel values
• Superscript symbols
• Meaningful comments
• Nested HTML tables

If that list makes your blood boil, you’ll enjoy the function names.

• behead() deals with multi-headered hydra tables one layer of headers at a time, working from the edge of the table inwards. It’s a bit like using header = TRUE in read.csv(), but because it’s a function, you can apply it to as many layers of headers as you need. You end up with all the headers in columns.
• spatter() is like tidyr::spread() but preserves mixed data types. You get into a mixed-data-type situation by delaying type coercion until after the table is tidy (rather than before, like read.csv() et al). And yes, it usually follows behead().

More positive, corrective functions:

• justify() aligns column headers before behead()ing, and has deliberate moral overtones.
• enhead() attaches a header to the body of the data, a la Frankenstein. The effect is the same as behead(), but is more powerful because you can choose exactly which header cells you want, paying attention to formatting (which behead() doesn’t understand).
• isolate_sentinels() separates meaningful symbols like "N/A" or "confidential" from the rest of the data, giving them some time alone think about what they’ve done.
• partition() takes a sheet with several tables on it, and slashes into pieces that each contain one table. You can then unpivot each table in turn with purrr::map() or similar.

Make cells tidy

Unpivotr uses data where each cells is represented by one row in a dataframe. Like this.

What can you do with tidy cells? The best places to start are:

• Spreadsheet Munging Strategies, a free, online cookbook using tidyxl and unpivotr
• Screencasts on YouTube.
• Worked examples on GitHub.

Otherwise the basic idea is:

1. Read the data with a specialist tool.
   • For spreadsheets, use tidyxl.
   • For plain text files, you might soon be able to use readr, but for now you’ll have to install a pull-request on that package with devtools::install_github("tidyverse/readr#760").
   • For tables in html pages, use unpivotr::tidy_html()
   • For data frames, use unpivotr::as_cells() – this should be a last resort, because by the time the data is in a conventional data frame, it is often too late – formatting has been lost, and most data types have been coerced to strings.
2. Either behead() straight away, else dplyr::filter() separately for the header cells and the data cells, and then recombine with enhead().
3. spatter() so that each column has one data type.

library(unpivotr)
library(tidyverse)
#> ── Attaching packages ────────────────────────────────── tidyverse 1.2.1 ──
#> ✔ ggplot2 3.1.0           ✔ purrr   0.2.5.9000 
#> ✔ tibble  1.4.99.9006     ✔ dplyr   0.7.8      
#> ✔ tidyr   0.8.2           ✔ stringr 1.3.1      
#> ✔ readr   1.2.1.9000      ✔ forcats 0.3.0
#> ── Conflicts ───────────────────────────────────── tidyverse_conflicts() ──
#> ✖ dplyr::filter() masks stats::filter()
#> ✖ dplyr::lag()    masks stats::lag()
x <- purpose$`NNW WNW`
x # A pivot table in a conventional data frame.  Four levels of headers, in two
#>                            X2      X3     X4     X5    X6     X7
#> 1                        <NA>    <NA> Female   <NA>  Male   <NA>
#> 2                        <NA>    <NA>  0 - 6 7 - 10 0 - 6 7 - 10
#> 3           Bachelor's degree 15 - 24   7000  27000  <NA>  13000
#> 4                        <NA> 25 - 44  12000 137000  9000  81000
#> 5                        <NA> 45 - 64  10000  64000  7000  66000
#> 6                        <NA>     65+   <NA>  18000  7000  17000
#> 7                 Certificate 15 - 24  29000 161000 30000 190000
#> 8                        <NA> 25 - 44  34000 179000 31000 219000
#> 9                        <NA> 45 - 64  30000 210000 23000 199000
#> 10                       <NA>     65+  12000  77000  8000 107000
#> 11                    Diploma 15 - 24   <NA>  14000  9000  11000
#> 12                       <NA> 25 - 44  10000  66000  8000  47000
#> 13                       <NA> 45 - 64   6000  68000  5000  58000
#> 14                       <NA>     65+   5000  41000  1000  34000
#> 15           No Qualification 15 - 24  10000  43000 12000  37000
#> 16                       <NA> 25 - 44  11000  36000 21000  50000
#> 17                       <NA> 45 - 64  19000  91000 17000  75000
#> 18                       <NA>     65+  16000 118000  9000  66000
#> 19 Postgraduate qualification 15 - 24   <NA>   6000  <NA>   <NA>
#> 20                       <NA> 25 - 44   5000  86000  7000  60000
#> 21                       <NA> 45 - 64   6000  55000  6000  68000
#> 22                       <NA>     65+   <NA>  13000  <NA>  18000
  # rows and two columns.
 
y <- as_cells(x) # 'Tokenize' or 'melt' the data frame into one row per cell
y
#> # A tibble: 132 x 4
#>      row   col data_type chr              
#>    <int> <int> <chr>     <chr>            
#>  1     1     1 chr       <NA>             
#>  2     2     1 chr       <NA>             
#>  3     3     1 chr       Bachelor's degree
#>  4     4     1 chr       <NA>             
#>  5     5     1 chr       <NA>             
#>  6     6     1 chr       <NA>             
#>  7     7     1 chr       Certificate      
#>  8     8     1 chr       <NA>             
#>  9     9     1 chr       <NA>             
#> 10    10     1 chr       <NA>             
#> # … with 122 more rows
 
rectify(y) # useful for reviewing the melted form as though in a spreadsheet
#> # A tibble: 22 x 7
#>    `row/col` `1(A)`            `2(B)`  `3(C)` `4(D)` `5(E)` `6(F)`
#>        <int> <chr>             <chr>   <chr>  <chr>  <chr>  <chr> 
#>  1         1 <NA>              <NA>    Female <NA>   Male   <NA>  
#>  2         2 <NA>              <NA>    0 - 6  7 - 10 0 - 6  7 - 10
#>  3         3 Bachelor's degree 15 - 24 7000   27000  <NA>   13000 
#>  4         4 <NA>              25 - 44 12000  137000 9000   81000 
#>  5         5 <NA>              45 - 64 10000  64000  7000   66000 
#>  6         6 <NA>              65+     <NA>   18000  7000   17000 
#>  7         7 Certificate       15 - 24 29000  161000 30000  190000
#>  8         8 <NA>              25 - 44 34000  179000 31000  219000
#>  9         9 <NA>              45 - 64 30000  210000 23000  199000
#> 10        10 <NA>              65+     12000  77000  8000   107000
#> # … with 12 more rows
 
y %>%
  behead("NNW", "sex") %>%               # Strip headers
  behead("N", "life-satisfication") %>%  # one
  behead("WNW", "qualification") %>%     # by
  behead("W", "age-band") %>%            # one.
  select(-row, -col, -data_type, count = chr) %>% # cleanup
  mutate(count = as.integer(count))
#> # A tibble: 80 x 5
#>     count sex    `life-satisfication` qualification     `age-band`
#>     <int> <chr>  <chr>                <chr>             <chr>     
#>  1   7000 Female 0 - 6                Bachelor's degree 15 - 24   
#>  2  12000 Female 0 - 6                Bachelor's degree 25 - 44   
#>  3  10000 Female 0 - 6                Bachelor's degree 45 - 64   
#>  4     NA Female 0 - 6                Bachelor's degree 65+       
#>  5  27000 Female 7 - 10               Bachelor's degree 15 - 24   
#>  6 137000 Female 7 - 10               Bachelor's degree 25 - 44   
#>  7  64000 Female 7 - 10               Bachelor's degree 45 - 64   
#>  8  18000 Female 7 - 10               Bachelor's degree 65+       
#>  9     NA Male   0 - 6                Bachelor's degree 15 - 24   
#> 10   9000 Male   0 - 6                Bachelor's degree 25 - 44   
#> # … with 70 more rows

Note the compass directions in the code above, which hint to behead() where to find the header cell for each data cell.

• "NNW" means the header (Female, Male) is positioned up and to the left of the columns of data cells it describes.
• "N" means the header (0 - 6, 7 - 10) is positioned directly above the columns of data cells it describes.
• "WNW" means the header (Bachelor's degree, Certificate, etc.) is positioned to the left and upwards of the rows of data cells it describes.
• "W" means the header (15 - 24, 25 - 44, etc.) is positioned directly to the left of the rows of data cells it describes.

Installation

# install.packages("devtools") # If you don't already have devtools
devtools::install_github("nacnudus/unpivotr", build_vignettes = TRUE)

The version 0.4.0 release had somee breaking changes. See NEWS.md for details. The previous version can be installed as follow:

devtools::install_version("unpivotr", version = "0.3.1", repos = "http://cran.us.r-project.org")

Similar projects

unpivotr is inspired by Databaker, a collaboration between the United Kingdom Office of National Statistics and The Sensible Code Company. unpivotr.

jailbreaker attempts to extract non-tabular data from spreadsheets into tabular structures automatically via some clever algorithms. unpivotr differs by being less magic, and equipping you to express what you want to do.