The Split-Apply-Combine Strategy
Many data analysis tasks involve three steps:
- splitting a data set into groups,
- applying some functions to each of the groups,
- combining the results.
Note that any of the steps 1 and 3 of this general procedure can be dropped, in which case we just transform a data frame without grouping it and later combining the result.
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 DataFrames package supports the split-apply-combine strategy through the groupby function that creates a GroupedDataFrame, followed by combine, select/select! or transform/transform!.
All operations described in this section of the manual are supported both for AbstractDataFrame (when split and combine steps are skipped) and GroupedDataFrame. Technically, AbstractDataFrame is just considered as being grouped on no columns (meaning it has a single group, or zero groups if it is empty). The only difference is that in this case the keepkeys and ungroup keyword arguments (described below) are not supported and a data frame is always returned, as there are no split and combine steps in this case.
In order to perform operations by groups you first need to create a GroupedDataFrame object from your data frame using the groupby function that takes two arguments: (1) a data frame to be grouped, and (2) a set of columns to group by.
Operations can then be applied on each group using one of the following functions:
combine: does not put restrictions on number of rows returned, the order of rows is specified by the order of groups inGroupedDataFrame; it is typically used to compute summary statistics by group;select: return a data frame with the number and order of rows exactly the same as the source data frame, including only new calculated columns;select!is an in-place version ofselect;transform: return a data frame with the number and order of rows exactly the same as the source data frame, including all columns from the source and new calculated columns;transform!is an in-place version oftransform.
All these functions take a specification of one or more functions to apply to each subset of the DataFrame. This specification can be of the following forms:
- standard column selectors (integers,
Symbols, strings, vectors of integers, vectors ofSymbols, vectors of strings,All,Cols,:,Between,Notand regular expressions) - a
cols => functionpair indicating thatfunctionshould be called with positional arguments holding columnscols, which can be any valid column selector; in this case target column name is automatically generated and it is assumed thatfunctionreturns a single value or a vector; the generated name is created by concatenating source column name andfunctionname by default (see examples below). - a
cols => function => target_colsform additionally explicitly specifying the target column or columns. - a
col => target_colspair, which renames the columncoltotarget_cols, which must be single name (as aSymbolor a string). - a
nrowornrow => target_colsform which efficiently computes the number of rows in a group; withouttarget_colsthe new column is called:nrow, otherwise it must be single name (as aSymbolor a string). - vectors or matrices containing transformations specified by the
Pairsyntax described in points 2 to 5 - a function which will be called with a
SubDataFramecorresponding to each group; this form should be avoided due to its poor performance unless the number of groups is small or a very large number of columns are processed (in which caseSubDataFrameavoids excessive compilation)
All functions have two types of signatures. One of them takes a GroupedDataFrame as the first argument and an arbitrary number of transformations described above as following arguments. The second type of signature is when a Function or a Type is passed as the first argument and a GroupedDataFrame as the second argument (similar to map).
As a special rule, with the cols => function and cols => function => target_cols syntaxes, if cols is wrapped in an AsTable object then a NamedTuple containing columns selected by cols is passed to function.
What is allowed for function to return is determined by the target_cols value:
- If both
colsandtarget_colsare omitted (so only afunctionis passed), then returning a data frame, a matrix, aNamedTuple, or aDataFrameRowwill produce multiple columns in the result. Returning any other value produces a single column. - If
target_colsis aSymbolor a string then the function is assumed to return a single column. In this case returning a data frame, a matrix, aNamedTuple, or aDataFrameRowraises an error. - If
target_colsis a vector ofSymbols or strings orAsTableit is assumed thatfunctionreturns multiple columns. Iffunctionreturns one ofAbstractDataFrame,NamedTuple,DataFrameRow,AbstractMatrixthen rules described in point 1 above apply. Iffunctionreturns anAbstractVectorthen each element of this vector must support thekeysfunction, which must return a collection ofSymbols, strings or integers; the return value ofkeysmust be identical for all elements. Then as many columns are created as there are elements in the return value of thekeysfunction. Iftarget_colsisAsTablethen their names are set to be equal to the key names except ifkeysreturns integers, in which case they are prefixed byx(so the column names are e.g.x1,x2, ...). Iftarget_colsis a vector ofSymbols or strings then column names produced using the rules above are ignored and replaced bytarget_cols(the number of columns must be the same as the length oftarget_colsin this case). Iffunreturns a value of any other type then it is assumed that it is a table conforming to the Tables.jl API and theTables.columntablefunction is called on it to get the resulting columns and their names. The names are retained whentarget_colsisAsTableand are replaced iftarget_colsis a vector ofSymbols or strings.
In all of these cases, function can return either a single row or multiple rows. As a particular rule, values wrapped in a Ref or a 0-dimensional AbstractArray are unwrapped and then treated as a single row.
select/select! and transform/transform! always return a DataFrame with the same number and order of rows as the source (even if GroupedDataFrame had its groups reordered).
For combine, rows in the returned object appear in the order of groups in the GroupedDataFrame. The functions can return an arbitrary number of rows for each group, but the kind of returned object and the number and names of columns must be the same for all groups, except when a DataFrame() or NamedTuple() is returned, in which case a given group is skipped.
It is allowed to mix single values and vectors if multiple transformations are requested. In this case single value will be repeated to match the length of columns specified by returned vectors.
A separate task is spawned for each specified transformation; each transformation then spawns as many tasks as Julia threads, and splits processing of groups across them (however, currently transformations with optimized implementations like sum and transformations that return multiple rows use a single task for all groups). This allows for parallel operation when Julia was started with more than one thread. Passed transformation functions should therefore not modify global variables (i.e. they should be pure), or use locks to control parallel accesses.
To apply function to each row instead of whole columns, it can be wrapped in a ByRow struct. cols can be any column indexing syntax, in which case function will be passed one argument for each of the columns specified by cols or a NamedTuple of them if specified columns are wrapped in AsTable. If ByRow is used it is allowed for cols to select an empty set of columns, in which case function is called for each row without any arguments and an empty NamedTuple is passed if empty set of columns is wrapped in AsTable.
The following keyword arguments are supported by the transformation functions (not all keyword arguments are supported in all cases; in general they are allowed in situations when they are meaningful, see the documentation of the specific functions for details):
keepkeys: whether grouping columns should be kept in the returned data frame.ungroup: whether the return value of the operation should be a data frame or aGroupedDataFrame.copycols: whether columns of the source data frame should be copied if no transformation is applied to them.renamecols: whether in thecols => functionform automatically generated column names should include the name of transformation functions or not.
We show several examples of these functions applied to the iris dataset below:
julia> using DataFrames, CSV, Statistics
julia> iris = CSV.read((joinpath(dirname(pathof(DataFrames)),
"..", "docs", "src", "assets", "iris.csv")),
DataFrame)
150×5 DataFrame
Row │ SepalLength SepalWidth PetalLength PetalWidth Species
│ Float64 Float64 Float64 Float64 String
─────┼──────────────────────────────────────────────────────────────────
1 │ 5.1 3.5 1.4 0.2 Iris-setosa
2 │ 4.9 3.0 1.4 0.2 Iris-setosa
3 │ 4.7 3.2 1.3 0.2 Iris-setosa
4 │ 4.6 3.1 1.5 0.2 Iris-setosa
5 │ 5.0 3.6 1.4 0.2 Iris-setosa
6 │ 5.4 3.9 1.7 0.4 Iris-setosa
7 │ 4.6 3.4 1.4 0.3 Iris-setosa
8 │ 5.0 3.4 1.5 0.2 Iris-setosa
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮
144 │ 6.8 3.2 5.9 2.3 Iris-virginica
145 │ 6.7 3.3 5.7 2.5 Iris-virginica
146 │ 6.7 3.0 5.2 2.3 Iris-virginica
147 │ 6.3 2.5 5.0 1.9 Iris-virginica
148 │ 6.5 3.0 5.2 2.0 Iris-virginica
149 │ 6.2 3.4 5.4 2.3 Iris-virginica
150 │ 5.9 3.0 5.1 1.8 Iris-virginica
135 rows omitted
julia> gdf = groupby(iris, :Species)
GroupedDataFrame with 3 groups based on key: Species
First Group (50 rows): Species = "Iris-setosa"
Row │ SepalLength SepalWidth PetalLength PetalWidth Species
│ Float64 Float64 Float64 Float64 String
─────┼───────────────────────────────────────────────────────────────
1 │ 5.1 3.5 1.4 0.2 Iris-setosa
2 │ 4.9 3.0 1.4 0.2 Iris-setosa
3 │ 4.7 3.2 1.3 0.2 Iris-setosa
4 │ 4.6 3.1 1.5 0.2 Iris-setosa
5 │ 5.0 3.6 1.4 0.2 Iris-setosa
6 │ 5.4 3.9 1.7 0.4 Iris-setosa
7 │ 4.6 3.4 1.4 0.3 Iris-setosa
8 │ 5.0 3.4 1.5 0.2 Iris-setosa
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮
43 │ 4.4 3.2 1.3 0.2 Iris-setosa
44 │ 5.0 3.5 1.6 0.6 Iris-setosa
45 │ 5.1 3.8 1.9 0.4 Iris-setosa
46 │ 4.8 3.0 1.4 0.3 Iris-setosa
47 │ 5.1 3.8 1.6 0.2 Iris-setosa
48 │ 4.6 3.2 1.4 0.2 Iris-setosa
49 │ 5.3 3.7 1.5 0.2 Iris-setosa
50 │ 5.0 3.3 1.4 0.2 Iris-setosa
34 rows omitted
⋮
Last Group (50 rows): Species = "Iris-virginica"
Row │ SepalLength SepalWidth PetalLength PetalWidth Species
│ Float64 Float64 Float64 Float64 String
─────┼──────────────────────────────────────────────────────────────────
1 │ 6.3 3.3 6.0 2.5 Iris-virginica
2 │ 5.8 2.7 5.1 1.9 Iris-virginica
3 │ 7.1 3.0 5.9 2.1 Iris-virginica
4 │ 6.3 2.9 5.6 1.8 Iris-virginica
5 │ 6.5 3.0 5.8 2.2 Iris-virginica
6 │ 7.6 3.0 6.6 2.1 Iris-virginica
7 │ 4.9 2.5 4.5 1.7 Iris-virginica
8 │ 7.3 2.9 6.3 1.8 Iris-virginica
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮
43 │ 5.8 2.7 5.1 1.9 Iris-virginica
44 │ 6.8 3.2 5.9 2.3 Iris-virginica
45 │ 6.7 3.3 5.7 2.5 Iris-virginica
46 │ 6.7 3.0 5.2 2.3 Iris-virginica
47 │ 6.3 2.5 5.0 1.9 Iris-virginica
48 │ 6.5 3.0 5.2 2.0 Iris-virginica
49 │ 6.2 3.4 5.4 2.3 Iris-virginica
50 │ 5.9 3.0 5.1 1.8 Iris-virginica
34 rows omitted
julia> combine(gdf, :PetalLength => mean)
3×2 DataFrame
Row │ Species PetalLength_mean
│ String Float64
─────┼───────────────────────────────────
1 │ Iris-setosa 1.464
2 │ Iris-versicolor 4.26
3 │ Iris-virginica 5.552
julia> combine(gdf, nrow)
3×2 DataFrame
Row │ Species nrow
│ String Int64
─────┼────────────────────────
1 │ Iris-setosa 50
2 │ Iris-versicolor 50
3 │ Iris-virginica 50
julia> combine(gdf, nrow, :PetalLength => mean => :mean)
3×3 DataFrame
Row │ Species nrow mean
│ String Int64 Float64
─────┼─────────────────────────────────
1 │ Iris-setosa 50 1.464
2 │ Iris-versicolor 50 4.26
3 │ Iris-virginica 50 5.552
julia> combine(gdf, [:PetalLength, :SepalLength] => ((p, s) -> (a=mean(p)/mean(s), b=sum(p))) =>
AsTable) # multiple columns are passed as arguments
3×3 DataFrame
Row │ Species a b
│ String Float64 Float64
─────┼────────────────────────────────────
1 │ Iris-setosa 0.292449 73.2
2 │ Iris-versicolor 0.717655 213.0
3 │ Iris-virginica 0.842744 277.6
julia> combine(gdf,
AsTable([:PetalLength, :SepalLength]) =>
x -> std(x.PetalLength) / std(x.SepalLength)) # passing a NamedTuple
3×2 DataFrame
Row │ Species PetalLength_SepalLength_function
│ String Float64
─────┼───────────────────────────────────────────────────
1 │ Iris-setosa 0.492245
2 │ Iris-versicolor 0.910378
3 │ Iris-virginica 0.867923
julia> combine(x -> std(x.PetalLength) / std(x.SepalLength), gdf) # passing a SubDataFrame
3×2 DataFrame
Row │ Species x1
│ String Float64
─────┼───────────────────────────
1 │ Iris-setosa 0.492245
2 │ Iris-versicolor 0.910378
3 │ Iris-virginica 0.867923
julia> combine(gdf, 1:2 => cor, nrow)
3×3 DataFrame
Row │ Species SepalLength_SepalWidth_cor nrow
│ String Float64 Int64
─────┼────────────────────────────────────────────────────
1 │ Iris-setosa 0.74678 50
2 │ Iris-versicolor 0.525911 50
3 │ Iris-virginica 0.457228 50
julia> combine(gdf, :PetalLength => (x -> [extrema(x)]) => [:min, :max])
3×3 DataFrame
Row │ Species min max
│ String Float64 Float64
─────┼───────────────────────────────────
1 │ Iris-setosa 1.0 1.9
2 │ Iris-versicolor 3.0 5.1
3 │ Iris-virginica 4.5 6.9Contrary to combine, the select and transform functions always return a data frame with the same number and order of rows as the source. In the example below the return values in columns :SepalLength_SepalWidth_cor and :nrow are broadcasted to match the number of elements in each group:
julia> select(gdf, 1:2 => cor)
150×2 DataFrame
Row │ Species SepalLength_SepalWidth_cor
│ String Float64
─────┼────────────────────────────────────────────
1 │ Iris-setosa 0.74678
2 │ Iris-setosa 0.74678
3 │ Iris-setosa 0.74678
4 │ Iris-setosa 0.74678
⋮ │ ⋮ ⋮
148 │ Iris-virginica 0.457228
149 │ Iris-virginica 0.457228
150 │ Iris-virginica 0.457228
143 rows omitted
julia> transform(gdf, :Species => x -> chop.(x, head=5, tail=0))
150×6 DataFrame
Row │ SepalLength SepalWidth PetalLength PetalWidth Species Species_function
│ Float64 Float64 Float64 Float64 String SubString…
─────┼────────────────────────────────────────────────────────────────────────────────────
1 │ 5.1 3.5 1.4 0.2 Iris-setosa setosa
2 │ 4.9 3.0 1.4 0.2 Iris-setosa setosa
3 │ 4.7 3.2 1.3 0.2 Iris-setosa setosa
4 │ 4.6 3.1 1.5 0.2 Iris-setosa setosa
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
148 │ 6.5 3.0 5.2 2.0 Iris-virginica virginica
149 │ 6.2 3.4 5.4 2.3 Iris-virginica virginica
150 │ 5.9 3.0 5.1 1.8 Iris-virginica virginica
143 rows omittedAll functions also support the do block form. However, as noted above, this form is slow and should therefore be avoided when performance matters.
julia> combine(gdf) do df
(m = mean(df.PetalLength), s² = var(df.PetalLength))
end
3×3 DataFrame
Row │ Species m s²
│ String Float64 Float64
─────┼─────────────────────────────────────
1 │ Iris-setosa 1.464 0.0301061
2 │ Iris-versicolor 4.26 0.220816
3 │ Iris-virginica 5.552 0.304588If you only want to split the data set into subsets, use the groupby function:
julia> for subdf in groupby(iris, :Species)
println(size(subdf, 1))
end
50
50
50To also get the values of the grouping columns along with each group, use the pairs function:
julia> for (key, subdf) in pairs(groupby(iris, :Species))
println("Number of data points for $(key.Species): $(nrow(subdf))")
end
Number of data points for Iris-setosa: 50
Number of data points for Iris-versicolor: 50
Number of data points for Iris-virginica: 50The value of key in the previous example is a DataFrames.GroupKey object, which can be used in a similar fashion to a NamedTuple.
Grouping a data frame using the groupby function can be seen as adding a lookup key to it. Such lookups can be performed efficiently by indexing the resulting GroupedDataFrame with a Tuple or NamedTuple:
julia> df = DataFrame(g = repeat(1:1000, inner=5), x = 1:5000)
5000×2 DataFrame
Row │ g x
│ Int64 Int64
──────┼──────────────
1 │ 1 1
2 │ 1 2
3 │ 1 3
4 │ 1 4
5 │ 1 5
6 │ 2 6
7 │ 2 7
8 │ 2 8
⋮ │ ⋮ ⋮
4994 │ 999 4994
4995 │ 999 4995
4996 │ 1000 4996
4997 │ 1000 4997
4998 │ 1000 4998
4999 │ 1000 4999
5000 │ 1000 5000
4985 rows omitted
julia> gdf = groupby(df, :g)
GroupedDataFrame with 1000 groups based on key: g
First Group (5 rows): g = 1
Row │ g x
│ Int64 Int64
─────┼──────────────
1 │ 1 1
2 │ 1 2
3 │ 1 3
4 │ 1 4
5 │ 1 5
⋮
Last Group (5 rows): g = 1000
Row │ g x
│ Int64 Int64
─────┼──────────────
1 │ 1000 4996
2 │ 1000 4997
3 │ 1000 4998
4 │ 1000 4999
5 │ 1000 5000
julia> gdf[(g=500,)]
5×2 SubDataFrame
Row │ g x
│ Int64 Int64
─────┼──────────────
1 │ 500 2496
2 │ 500 2497
3 │ 500 2498
4 │ 500 2499
5 │ 500 2500
julia> gdf[[(500,), (501,)]]
GroupedDataFrame with 2 groups based on key: g
First Group (5 rows): g = 500
Row │ g x
│ Int64 Int64
─────┼──────────────
1 │ 500 2496
2 │ 500 2497
3 │ 500 2498
4 │ 500 2499
5 │ 500 2500
⋮
Last Group (5 rows): g = 501
Row │ g x
│ Int64 Int64
─────┼──────────────
1 │ 501 2501
2 │ 501 2502
3 │ 501 2503
4 │ 501 2504
5 │ 501 2505In order to apply a function to each non-grouping column of a GroupedDataFrame you can write:
julia> gd = groupby(iris, :Species)
GroupedDataFrame with 3 groups based on key: Species
First Group (50 rows): Species = "Iris-setosa"
Row │ SepalLength SepalWidth PetalLength PetalWidth Species
│ Float64 Float64 Float64 Float64 String
─────┼───────────────────────────────────────────────────────────────
1 │ 5.1 3.5 1.4 0.2 Iris-setosa
2 │ 4.9 3.0 1.4 0.2 Iris-setosa
3 │ 4.7 3.2 1.3 0.2 Iris-setosa
4 │ 4.6 3.1 1.5 0.2 Iris-setosa
5 │ 5.0 3.6 1.4 0.2 Iris-setosa
6 │ 5.4 3.9 1.7 0.4 Iris-setosa
7 │ 4.6 3.4 1.4 0.3 Iris-setosa
8 │ 5.0 3.4 1.5 0.2 Iris-setosa
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮
43 │ 4.4 3.2 1.3 0.2 Iris-setosa
44 │ 5.0 3.5 1.6 0.6 Iris-setosa
45 │ 5.1 3.8 1.9 0.4 Iris-setosa
46 │ 4.8 3.0 1.4 0.3 Iris-setosa
47 │ 5.1 3.8 1.6 0.2 Iris-setosa
48 │ 4.6 3.2 1.4 0.2 Iris-setosa
49 │ 5.3 3.7 1.5 0.2 Iris-setosa
50 │ 5.0 3.3 1.4 0.2 Iris-setosa
34 rows omitted
⋮
Last Group (50 rows): Species = "Iris-virginica"
Row │ SepalLength SepalWidth PetalLength PetalWidth Species
│ Float64 Float64 Float64 Float64 String
─────┼──────────────────────────────────────────────────────────────────
1 │ 6.3 3.3 6.0 2.5 Iris-virginica
2 │ 5.8 2.7 5.1 1.9 Iris-virginica
3 │ 7.1 3.0 5.9 2.1 Iris-virginica
4 │ 6.3 2.9 5.6 1.8 Iris-virginica
5 │ 6.5 3.0 5.8 2.2 Iris-virginica
6 │ 7.6 3.0 6.6 2.1 Iris-virginica
7 │ 4.9 2.5 4.5 1.7 Iris-virginica
8 │ 7.3 2.9 6.3 1.8 Iris-virginica
⋮ │ ⋮ ⋮ ⋮ ⋮ ⋮
43 │ 5.8 2.7 5.1 1.9 Iris-virginica
44 │ 6.8 3.2 5.9 2.3 Iris-virginica
45 │ 6.7 3.3 5.7 2.5 Iris-virginica
46 │ 6.7 3.0 5.2 2.3 Iris-virginica
47 │ 6.3 2.5 5.0 1.9 Iris-virginica
48 │ 6.5 3.0 5.2 2.0 Iris-virginica
49 │ 6.2 3.4 5.4 2.3 Iris-virginica
50 │ 5.9 3.0 5.1 1.8 Iris-virginica
34 rows omitted
julia> combine(gd, valuecols(gd) .=> mean)
3×5 DataFrame
Row │ Species SepalLength_mean SepalWidth_mean PetalLength_mean P ⋯
│ String Float64 Float64 Float64 F ⋯
─────┼──────────────────────────────────────────────────────────────────────────
1 │ Iris-setosa 5.006 3.418 1.464 ⋯
2 │ Iris-versicolor 5.936 2.77 4.26
3 │ Iris-virginica 6.588 2.974 5.552
1 column omitted