Working with Data Frames
Examining the Data
The default printing of DataFrame objects only includes a sample of rows and columns that fits on screen:
julia> using DataFrames
julia> df = DataFrame(A = 1:2:1000, B = repeat(1:10, inner=50), C = 1:500)
500×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 3 1 2
3 │ 5 1 3
4 │ 7 1 4
5 │ 9 1 5
6 │ 11 1 6
7 │ 13 1 7
8 │ 15 1 8
⋮ │ ⋮ ⋮ ⋮
494 │ 987 10 494
495 │ 989 10 495
496 │ 991 10 496
497 │ 993 10 497
498 │ 995 10 498
499 │ 997 10 499
500 │ 999 10 500
485 rows omittedPrinting options can be adjusted by calling the show function manually: show(df, allrows=true) prints all rows even if they do not fit on screen and show(df, allcols=true) does the same for columns.
The first and last functions can be used to look at the first and last rows of a data frame (respectively):
julia> first(df, 6)
6×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 3 1 2
3 │ 5 1 3
4 │ 7 1 4
5 │ 9 1 5
6 │ 11 1 6
julia> last(df, 6)
6×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 989 10 495
2 │ 991 10 496
3 │ 993 10 497
4 │ 995 10 498
5 │ 997 10 499
6 │ 999 10 500Also notice that when DataFrame is printed to the console or rendered in HTML (e.g. in Jupyter Notebook) you get an information about type of elements held in its columns. For example in this case:
julia> using CategoricalArrays
julia> DataFrame(a = 1:2, b = [1.0, missing],
c = categorical('a':'b'), d = [1//2, missing])
2×4 DataFrame
Row │ a b c d
│ Int64 Float64? Cat… Rational…?
─────┼────────────────────────────────────
1 │ 1 1.0 a 1//2
2 │ 2 missing b missing
we can observe that:
- the first column
:acan hold elements of typeInt64; - the second column
:bcan holdFloat64orMissing, which is indicated by?printed after the name of type; - the third column
:ccan hold categorical data; here we notice…, which indicates that the actual name of the type was long and got truncated; - the type information in fourth column
:dpresents a situation where the name is both truncated and the type allowsMissing.
Taking a Subset
Indexing syntax
Specific subsets of a data frame can be extracted using the indexing syntax, similar to matrices. In the Indexing section of the manual you can find all the details about the available options. Here we highlight the basic options.
The colon : indicates that all items (rows or columns depending on its position) should be retained:
julia> df[1:3, :]
3×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 3 1 2
3 │ 5 1 3
julia> df[[1, 5, 10], :]
3×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 9 1 5
3 │ 19 1 10
julia> df[:, [:A, :B]]
500×2 DataFrame
Row │ A B
│ Int64 Int64
─────┼──────────────
1 │ 1 1
2 │ 3 1
3 │ 5 1
4 │ 7 1
5 │ 9 1
6 │ 11 1
7 │ 13 1
8 │ 15 1
⋮ │ ⋮ ⋮
494 │ 987 10
495 │ 989 10
496 │ 991 10
497 │ 993 10
498 │ 995 10
499 │ 997 10
500 │ 999 10
485 rows omitted
julia> df[1:3, [:B, :A]]
3×2 DataFrame
Row │ B A
│ Int64 Int64
─────┼──────────────
1 │ 1 1
2 │ 1 3
3 │ 1 5
julia> df[[3, 1], [:C]]
2×1 DataFrame
Row │ C
│ Int64
─────┼───────
1 │ 3
2 │ 1Do note that df[!, [:A]] and df[:, [:A]] return a DataFrame object, while df[!, :A] and df[:, :A] return a vector:
julia> df[!, [:A]]
500×1 DataFrame
Row │ A
│ Int64
─────┼───────
1 │ 1
2 │ 3
3 │ 5
4 │ 7
5 │ 9
6 │ 11
7 │ 13
8 │ 15
⋮ │ ⋮
494 │ 987
495 │ 989
496 │ 991
497 │ 993
498 │ 995
499 │ 997
500 │ 999
485 rows omitted
julia> df[!, [:A]] == df[:, [:A]]
true
julia> df[!, :A]
500-element Vector{Int64}:
1
3
5
7
9
11
13
15
17
19
⋮
983
985
987
989
991
993
995
997
999
julia> df[!, :A] == df[:, :A]
trueIn the first case, [:A] is a vector, indicating that the resulting object should be a DataFrame. On the other hand, :A is a single symbol, indicating that a single column vector should be extracted. Note that in the first case a vector is required to be passed (not just any iterable), so e.g. df[:, (:x1, :x2)] is not allowed, but df[:, [:x1, :x2]] is valid.
It is also possible to use a regular expression as a selector of columns matching it:
julia> df = DataFrame(x1=1, x2=2, y=3)
1×3 DataFrame
Row │ x1 x2 y
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 2 3
julia> df[!, r"x"]
1×2 DataFrame
Row │ x1 x2
│ Int64 Int64
─────┼──────────────
1 │ 1 2A Not selector (from the InvertedIndices package) can be used to select all columns excluding a specific subset:
julia> df[!, Not(:x1)]
1×2 DataFrame
Row │ x2 y
│ Int64 Int64
─────┼──────────────
1 │ 2 3Finally, you can use Not, Between, Cols and All selectors in more complex column selection scenarios (note that Cols() selects no columns while All() selects all columns therefore Cols is a preferred selector if you write generic code). The following examples move all columns whose names match r"x" regular expression respectively to the front and to the end of a data frame:
julia> df = DataFrame(r=1, x1=2, x2=3, y=4)
1×4 DataFrame
Row │ r x1 x2 y
│ Int64 Int64 Int64 Int64
─────┼────────────────────────────
1 │ 1 2 3 4
julia> df[:, Cols(r"x", :)]
1×4 DataFrame
Row │ x1 x2 r y
│ Int64 Int64 Int64 Int64
─────┼────────────────────────────
1 │ 2 3 1 4
julia> df[:, Cols(Not(r"x"), :)]
1×4 DataFrame
Row │ r y x1 x2
│ Int64 Int64 Int64 Int64
─────┼────────────────────────────
1 │ 1 4 2 3The indexing syntax can also be used to select rows based on conditions on variables:
julia> df = DataFrame(A = 1:2:1000, B = repeat(1:10, inner=50), C = 1:500)
500×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 3 1 2
3 │ 5 1 3
4 │ 7 1 4
5 │ 9 1 5
6 │ 11 1 6
7 │ 13 1 7
8 │ 15 1 8
⋮ │ ⋮ ⋮ ⋮
494 │ 987 10 494
495 │ 989 10 495
496 │ 991 10 496
497 │ 993 10 497
498 │ 995 10 498
499 │ 997 10 499
500 │ 999 10 500
485 rows omitted
julia> df[df.A .> 500, :]
250×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 501 6 251
2 │ 503 6 252
3 │ 505 6 253
4 │ 507 6 254
5 │ 509 6 255
6 │ 511 6 256
7 │ 513 6 257
8 │ 515 6 258
⋮ │ ⋮ ⋮ ⋮
244 │ 987 10 494
245 │ 989 10 495
246 │ 991 10 496
247 │ 993 10 497
248 │ 995 10 498
249 │ 997 10 499
250 │ 999 10 500
235 rows omitted
julia> df[(df.A .> 500) .& (300 .< df.C .< 400), :]
99×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 601 7 301
2 │ 603 7 302
3 │ 605 7 303
4 │ 607 7 304
5 │ 609 7 305
6 │ 611 7 306
7 │ 613 7 307
8 │ 615 7 308
⋮ │ ⋮ ⋮ ⋮
93 │ 785 8 393
94 │ 787 8 394
95 │ 789 8 395
96 │ 791 8 396
97 │ 793 8 397
98 │ 795 8 398
99 │ 797 8 399
84 rows omittedWhere a specific subset of values needs to be matched, the in() function can be applied:
julia> df[in.(df.A, Ref([1, 5, 601])), :]
3×3 DataFrame
Row │ A B C
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 1 1
2 │ 5 1 3
3 │ 601 7 301Equivalently, the in function can be called with a single argument to create a function object that tests whether each value belongs to the subset (partial application of in): df[in([1, 5, 601]).(df.A), :].
As with matrices, subsetting from a data frame will usually return a copy of columns, not a view or direct reference.
The only indexing situations where data frames will not return a copy are:
- when a
!is placed in the first indexing position (df[!, :A], ordf[!, [:A, :B]]), - when using
.(getpropery) notation (df.A), - when a single row is selected using an integer (
df[1, [:A, :B]]) - when
viewor@viewis used (e.g.@view df[1:3, :A]).
More details on copies, views, and references can be found in the getindex and view section.
Selecting and transforming columns
You can also use the select/select! and transform/transform! functions to select, rename and transform columns in a data frame.
The select function creates a new data frame:
julia> df = DataFrame(x1=[1, 2], x2=[3, 4], y=[5, 6])
2×3 DataFrame
Row │ x1 x2 y
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 3 5
2 │ 2 4 6
julia> select(df, Not(:x1)) # drop column :x1 in a new data frame
2×2 DataFrame
Row │ x2 y
│ Int64 Int64
─────┼──────────────
1 │ 3 5
2 │ 4 6
julia> select(df, r"x") # select columns containing 'x' character
2×2 DataFrame
Row │ x1 x2
│ Int64 Int64
─────┼──────────────
1 │ 1 3
2 │ 2 4
julia> select(df, :x1 => :a1, :x2 => :a2) # rename columns
2×2 DataFrame
Row │ a1 a2
│ Int64 Int64
─────┼──────────────
1 │ 1 3
2 │ 2 4
julia> select(df, :x1, :x2 => (x -> x .- minimum(x)) => :x2) # transform columns
2×2 DataFrame
Row │ x1 x2
│ Int64 Int64
─────┼──────────────
1 │ 1 0
2 │ 2 1
julia> select(df, :x2, :x2 => ByRow(sqrt)) # transform columns by row
2×2 DataFrame
Row │ x2 x2_sqrt
│ Int64 Float64
─────┼────────────────
1 │ 3 1.73205
2 │ 4 2.0
julia> select(df, AsTable(:) => ByRow(extrema) => [:lo, :hi]) # return multiple columns
2×2 DataFrame
Row │ lo hi
│ Int64 Int64
─────┼──────────────
1 │ 1 5
2 │ 2 6It is important to note that select always returns a data frame, even if a single column is selected (as opposed to indexing syntax).
julia> select(df, :x1)
2×1 DataFrame
Row │ x1
│ Int64
─────┼───────
1 │ 1
2 │ 2
julia> df[:, :x1]
2-element Vector{Int64}:
1
2By default select copies columns of a passed source data frame. In order to avoid copying, pass copycols=false:
julia> df2 = select(df, :x1)
2×1 DataFrame
Row │ x1
│ Int64
─────┼───────
1 │ 1
2 │ 2
julia> df2.x1 === df.x1
false
julia> df2 = select(df, :x1, copycols=false)
2×1 DataFrame
Row │ x1
│ Int64
─────┼───────
1 │ 1
2 │ 2
julia> df2.x1 === df.x1
trueTo perform the selection operation in-place use select!:
julia> select!(df, Not(:x1));
julia> df
2×2 DataFrame
Row │ x2 y
│ Int64 Int64
─────┼──────────────
1 │ 3 5
2 │ 4 6transform and transform! functions work identically to select and select! with the only difference that they retain all columns that are present in the source data frame. Here are some more advanced examples.
First we show how to generate a column that is a sum of all other columns in the data frame using the All() selector:
julia> df = DataFrame(x1=[1, 2], x2=[3, 4], y=[5, 6])
2×3 DataFrame
Row │ x1 x2 y
│ Int64 Int64 Int64
─────┼─────────────────────
1 │ 1 3 5
2 │ 2 4 6
julia> transform(df, All() => +)
2×4 DataFrame
Row │ x1 x2 y x1_x2_y_+
│ Int64 Int64 Int64 Int64
─────┼────────────────────────────────
1 │ 1 3 5 9
2 │ 2 4 6 12Using the ByRow wrapper, we can easily compute for each row the name of column with the highest score:
julia> using Random
julia> Random.seed!(1);
julia> df = DataFrame(rand(10, 3), [:a, :b, :c])
10×3 DataFrame
Row │ a b c
│ Float64 Float64 Float64
─────┼──────────────────────────────────
1 │ 0.236033 0.555751 0.0769509
2 │ 0.346517 0.437108 0.640396
3 │ 0.312707 0.424718 0.873544
4 │ 0.00790928 0.773223 0.278582
5 │ 0.488613 0.28119 0.751313
6 │ 0.210968 0.209472 0.644883
7 │ 0.951916 0.251379 0.0778264
8 │ 0.999905 0.0203749 0.848185
9 │ 0.251662 0.287702 0.0856352
10 │ 0.986666 0.859512 0.553206
julia> transform(df, AsTable(:) => ByRow(argmax) => :prediction)
10×4 DataFrame
Row │ a b c prediction
│ Float64 Float64 Float64 Symbol
─────┼──────────────────────────────────────────────
1 │ 0.236033 0.555751 0.0769509 b
2 │ 0.346517 0.437108 0.640396 c
3 │ 0.312707 0.424718 0.873544 c
4 │ 0.00790928 0.773223 0.278582 b
5 │ 0.488613 0.28119 0.751313 c
6 │ 0.210968 0.209472 0.644883 c
7 │ 0.951916 0.251379 0.0778264 a
8 │ 0.999905 0.0203749 0.848185 a
9 │ 0.251662 0.287702 0.0856352 b
10 │ 0.986666 0.859512 0.553206 aIn the following, most complex, example below we compute row-wise sum, number of elements, and mean, while ignoring missing values.
julia> using Statistics
julia> df = DataFrame(x=[1, 2, missing], y=[1, missing, missing])
3×2 DataFrame
Row │ x y
│ Int64? Int64?
─────┼──────────────────
1 │ 1 1
2 │ 2 missing
3 │ missing missing
julia> transform(df, AsTable(:) .=>
ByRow.([sum∘skipmissing,
x -> count(!ismissing, x),
mean∘skipmissing]) .=>
[:sum, :n, :mean])
3×5 DataFrame
Row │ x y sum n mean
│ Int64? Int64? Int64 Int64 Float64
─────┼─────────────────────────────────────────
1 │ 1 1 2 2 1.0
2 │ 2 missing 2 1 2.0
3 │ missing missing 0 0 NaNWhile the DataFrames.jl package provides basic data manipulation capabilities, users are encouraged to use querying frameworks for more convenient and powerful operations:
- the Query.jl package provides a
LINQ-like interface to a large number of data sources
package provides interfaces similar to LINQ and dplyr
See the Data manipulation frameworks section for more information.
Summarizing Data
The describe function returns a data frame summarizing the elementary statistics and information about each column:
julia> df = DataFrame(A = 1:4, B = ["M", "F", "F", "M"])
4×2 DataFrame
Row │ A B
│ Int64 String
─────┼───────────────
1 │ 1 M
2 │ 2 F
3 │ 3 F
4 │ 4 M
julia> describe(df)
2×7 DataFrame
Row │ variable mean min median max nmissing eltype
│ Symbol Union… Any Union… Any Int64 DataType
─────┼────────────────────────────────────────────────────────
1 │ A 2.5 1 2.5 4 0 Int64
2 │ B F M 0 StringIf you are interested in describing only a subset of columns then the easiest way to do it is to pass a subset of an original data frame to describe like this:
julia> describe(df[!, [:A]])
1×7 DataFrame
Row │ variable mean min median max nmissing eltype
│ Symbol Float64 Int64 Float64 Int64 Int64 DataType
─────┼──────────────────────────────────────────────────────────────
1 │ A 2.5 1 2.5 4 0 Int64Of course, one can also compute descriptive statistics directly on individual columns:
julia> using Statistics
julia> mean(df.A)
2.5We can also apply a function to each column of a DataFrame using combine. For example:
julia> df = DataFrame(A = 1:4, B = 4.0:-1.0:1.0)
4×2 DataFrame
Row │ A B
│ Int64 Float64
─────┼────────────────
1 │ 1 4.0
2 │ 2 3.0
3 │ 3 2.0
4 │ 4 1.0
julia> combine(df, names(df) .=> sum)
1×2 DataFrame
Row │ A_sum B_sum
│ Int64 Float64
─────┼────────────────
1 │ 10 10.0
julia> combine(df, names(df) .=> sum, names(df) .=> prod)
1×4 DataFrame
Row │ A_sum B_sum A_prod B_prod
│ Int64 Float64 Int64 Float64
─────┼─────────────────────────────────
1 │ 10 10.0 24 24.0If you would prefer the result to have the same number of rows as the source data frame use select instead of combine.
Handling of Columns Stored in a DataFrame
Functions that transform a DataFrame to produce a new DataFrame always perform a copy of the columns by default, for example:
julia> df = DataFrame(A = 1:4, B = 4.0:-1.0:1.0)
4×2 DataFrame
Row │ A B
│ Int64 Float64
─────┼────────────────
1 │ 1 4.0
2 │ 2 3.0
3 │ 3 2.0
4 │ 4 1.0
julia> df2 = copy(df);
julia> df2.A === df.A
falseOn the other hand, in-place functions, whose names end with !, may mutate the column vectors of the DataFrame they take as an argument, for example:
julia> x = [3, 1, 2];
julia> df = DataFrame(x=x)
3×1 DataFrame
Row │ x
│ Int64
─────┼───────
1 │ 3
2 │ 1
3 │ 2
julia> sort!(df)
3×1 DataFrame
Row │ x
│ Int64
─────┼───────
1 │ 1
2 │ 2
3 │ 3
julia> x
3-element Vector{Int64}:
3
1
2
julia> df.x[1] = 100
100
julia> df
3×1 DataFrame
Row │ x
│ Int64
─────┼───────
1 │ 100
2 │ 2
3 │ 3
julia> x
3-element Vector{Int64}:
3
1
2Note, that in the above example the original x vector is not mutated in the process as the DataFrame(x=x) constructor makes a copy by default.
In-place functions are safe to call, except when a view of the DataFrame (created via a view, @view or groupby) or when a DataFrame created with copycols=false are in use.
It is possible to have a direct access to a column col of a DataFrame df using the syntaxes df.col, df[!, :col], via the eachcol function, by accessing a parent of a view of a column of a DataFrame, or simply by storing the reference to the column vector before the DataFrame was created with copycols=false.
julia> x = [3, 1, 2];
julia> df = DataFrame(x=x)
3×1 DataFrame
Row │ x
│ Int64
─────┼───────
1 │ 3
2 │ 1
3 │ 2
julia> df.x == x
true
julia> df[!, 1] !== x
true
julia> eachcol(df)[1] === df.x
trueNote that a column obtained from a DataFrame using one of these methods should not be mutated without caution.
The exact rules of handling columns of a DataFrame are explained in The design of handling of columns of a DataFrame section of the manual.
Replacing Data
Several approaches can be used to replace some values with others in a data frame. Some apply the replacement to all values in a data frame, and others to individual columns or subset of columns.
Do note that in-place replacement requires that the replacement value can be converted to the column's element type. In particular, this implies that replacing a value with missing requires a call to allowmissing! if the column did not allow for missing values.
Replacement operations affecting a single column can be performed using replace!:
julia> using DataFrames
julia> df = DataFrame(a = ["a", "None", "b", "None"], b = 1:4, c = ["None", "j", "k", "h"], d = ["x", "y", "None", "z"])
4×4 DataFrame
Row │ a b c d
│ String Int64 String String
─────┼───────────────────────────────
1 │ a 1 None x
2 │ None 2 j y
3 │ b 3 k None
4 │ None 4 h z
julia> replace!(df.a, "None" => "c")
4-element Vector{String}:
"a"
"c"
"b"
"c"
julia> df
4×4 DataFrame
Row │ a b c d
│ String Int64 String String
─────┼───────────────────────────────
1 │ a 1 None x
2 │ c 2 j y
3 │ b 3 k None
4 │ c 4 h zThis is equivalent to df.a = replace(df.a, "None" => "c"), but operates in-place, without allocating a new column vector.
Replacement operations on multiple columns or on the whole data frame can be performed in-place using the broadcasting syntax:
# replacement on a subset of columns [:c, :d]
julia> df[:, [:c, :d]] .= ifelse.(df[!, [:c, :d]] .== "None", "c", df[!, [:c, :d]])
4×2 SubDataFrame
Row │ c d
│ String String
─────┼────────────────
1 │ c x
2 │ j y
3 │ k c
4 │ h z
julia> df
4×4 DataFrame
Row │ a b c d
│ String Int64 String String
─────┼───────────────────────────────
1 │ a 1 c x
2 │ c 2 j y
3 │ b 3 k c
4 │ c 4 h z
julia> df .= ifelse.(df .== "c", "None", df) # replacement on entire data frame
4×4 DataFrame
Row │ a b c d
│ String Int64 String String
─────┼───────────────────────────────
1 │ a 1 None x
2 │ None 2 j y
3 │ b 3 k None
4 │ None 4 h zDo note that in the above examples, changing .= to just = will allocate new column vectors instead of applying the operation in-place.
When replacing values with missing, if the columns do not already allow for missing values, one has to either avoid in-place operation and use = instead of .=, or call allowmissing! beforehand:
julia> df2 = ifelse.(df .== "None", missing, df) # do not operate in-place (`df = ` would also work)
4×4 DataFrame
Row │ a b c d
│ String? Int64 String? String?
─────┼──────────────────────────────────
1 │ a 1 missing x
2 │ missing 2 j y
3 │ b 3 k missing
4 │ missing 4 h z
julia> allowmissing!(df) # operate in-place after allowing for missing
4×4 DataFrame
Row │ a b c d
│ String? Int64? String? String?
─────┼───────────────────────────────────
1 │ a 1 None x
2 │ None 2 j y
3 │ b 3 k None
4 │ None 4 h z
julia> df .= ifelse.(df .== "None", missing, df)
4×4 DataFrame
Row │ a b c d
│ String? Int64? String? String?
─────┼───────────────────────────────────
1 │ a 1 missing x
2 │ missing 2 j y
3 │ b 3 k missing
4 │ missing 4 h z