The Split-Apply-Combine Strategy

Many data analysis tasks involve splitting a data set into groups, applying some functions to each of the groups and then combining the results. A standardized framework for handling this sort of computation is described in the paper "The Split-Apply-Combine Strategy for Data Analysis", written by Hadley Wickham.

The DataTables package supports the Split-Apply-Combine strategy through the by function, which takes in three arguments: (1) a DataTable, (2) one or more columns to split the DataTable on, and (3) a function or expression to apply to each subset of the DataTable.

We show several examples of the by function applied to the iris dataset below:

using DataTables
using CSV
iris = CSV.read(joinpath(Pkg.dir("DataTables"), "test/data/iris.csv"), DataTable)

by(iris, :Species, size)
by(iris, :Species, dt -> mean(dropnull(dt[:PetalLength])))
by(iris, :Species, dt -> DataTable(N = size(dt, 1)))

The by function also support the do block form:

by(iris, :Species) do dt
   DataTable(m = mean(dropnull(dt[:PetalLength])), s² = var(dropnull(dt[:PetalLength])))
end

A second approach to the Split-Apply-Combine strategy is implemented in the aggregate function, which also takes three arguments: (1) a DataTable, (2) one or more columns to split the DataTable on, and (3) one or more functions that are used to compute a summary of each subset of the DataTable. Each function is applied to each column, that was not used to split the DataTable, creating new columns of the form $name_$function e.g. SepalLength_mean. Anonymous functions and expressions that do not have a name will be called λ1.

We show several examples of the aggregate function applied to the iris dataset below:

aggregate(iris, :Species, sum)
aggregate(iris, :Species, [sum, x->mean(dropnull(x))])

If you only want to split the data set into subsets, use the groupby function:

for subdt in groupby(iris, :Species)
    println(size(subdt, 1))
end