In patperry/r-frame: Flexible Data Frames

The package is built around the idea that a data point consists of a (variable, key, value) triple identifying an attribute, target, and value. This notion of data differs from Wickham's notion of "tidy" data, which allows only (variable, value) pairs. Having explicit support for keys makes it easier to link different measurements made on the same set of individuals and makes it easier to identify the sources giving rise to downstream results. R data.frame objects have partial support for keys through their rownames; the dataset object extends this support by allowing non-character and multi-component keys.

Grouping

Split the rows according to groups defined by one or more columns, optionally performing a computation on each group.

# split the rows into groups defined by unique ('cyl', 'gear') combinations;
# the grouping factors are the keys for the result
xg <- group(mtcars, cyl, gear)

# perform a computation on all groups
do(xg, function(x)
   record(n   = nrow(x),
          mpg = mean(x$mpg),
          hp  = mean(x$hp)))
#> cyl gear │  n    mpg       hp
#>   6    4 │  4 19.750 116.5000
#>   4    4 │  8 26.925  76.0000
#>   6    3 │  2 19.750 107.5000
#>   8    3 │ 12 15.050 194.1667
#>   4    3 │  1 21.500  97.0000
#>   4    5 │  2 28.200 102.0000
#>   8    5 │  2 15.400 299.5000
#>   6    5 │  1 19.700 175.0000