Possible pitfalls

using DataFrames
using BenchmarkTools

Know what is copied when creating a DataFrame

x = DataFrame(rand(3, 5), :auto)

3×5 DataFrame

Row	x1	x2	x3	x4	x5
	Float64	Float64	Float64	Float64	Float64
1	0.213634	0.976066	0.764957	0.23689	0.586611
2	0.257904	0.727199	0.125993	0.175773	0.556361
3	0.567658	0.580706	0.227893	0.628414	0.325745

x and y are not the same object

y = copy(x)
x === y

false

x and y are not the same object

y = DataFrame(x)
x === y

false

the columns are also not the same

any(x[!, i] === y[!, i] for i in ncol(x))

false

x and y are not the same object

y = DataFrame(x, copycols=false)
x === y

false

But the columns are the same

all(x[!, i] === y[!, i] for i in ncol(x))

true

the same when creating data frames using kwarg syntax

x = 1:3;
y = [1, 2, 3];
df = DataFrame(x=x, y=y)

3×2 DataFrame

Row	x	y
	Int64	Int64
1	1	1
2	2	2
3	3	3

different object

y === df.y

false

range is converted to a vector

typeof(x), typeof(df.x)

(UnitRange{Int64}, Vector{Int64})

slicing rows always creates a copy

y === df[:, :y]

false

you can avoid copying by using copycols=false keyword argument in functions.

df = DataFrame(x=x, y=y, copycols=false)

3×2 DataFrame

Row	x	y
	Int64	Int64
1	1	1
2	2	2
3	3	3

now it is the same

y === df.y

true

not the same object

select(df, :y)[!, 1] === y

false

the same object

select(df, :y, copycols=false)[!, 1] === y

true

Do not modify the parent of GroupedDataFrame or view

x = DataFrame(id=repeat([1, 2], outer=3), x=1:6)
g = groupby(x, :id)

x[1:3, 1] = [2, 2, 2]
g ## well - it is wrong now, g is only a view

GroupedDataFrame with 2 groups based on key: id

First Group (3 rows): id = 2

Row	id	x
	Int64	Int64
1	2	1
2	2	3
3	1	5

⋮

Last Group (3 rows): id = 2

Row	id	x
	Int64	Int64
1	2	2
2	2	4
3	2	6

s = view(x, 5:6, :)

2×2 SubDataFrame

Row	id	x
	Int64	Int64
1	1	5
2	2	6

delete!(x, 3:6)

2×2 DataFrame

Row	id	x
	Int64	Int64
1	2	1
2	2	2

This is an error

s ## Will return BoundsError

Single column selection for a DataFrame

Single column selection for a DataFrame creates aliases with ! and getproperty syntax and copies with :

x = DataFrame(a=1:3)
x.b = x[!, 1] ## alias
x.c = x[:, 1] ## copy
x.d = x[!, 1][:] ## copy
x.e = copy(x[!, 1]) ## explicit copy
display(x)

3×5 DataFrame

Row	a	b	c	d	e
	Int64	Int64	Int64	Int64	Int64
1	1	1	1	1	1
2	2	2	2	2	2
3	3	3	3	3	3

x[1, 1] = 100
display(x)

3×5 DataFrame

Row	a	b	c	d	e
	Int64	Int64	Int64	Int64	Int64
1	100	100	1	1	1
2	2	2	2	2	2
3	3	3	3	3	3

When iterating rows of a data frame

• use eachrow to avoid compilation cost in wide tables,

• but Tables.namedtupleiterator for fast execution in tall tables

The table below is tall:

df2 = DataFrame(rand(10^6, 10), :auto)

1000000×10 DataFrame

999975 rows omitted

Row	x1	x2	x3	x4	x5	x6	x7	x8	x9	x10
	Float64	Float64	Float64	Float64	Float64	Float64	Float64	Float64	Float64	Float64
1	0.22214	0.194725	0.787694	0.79179	0.577703	0.148028	0.090797	0.518032	0.430776	0.820852
2	0.797788	0.482724	0.104816	0.98981	0.280418	0.973437	0.171954	0.66921	0.489958	0.907005
3	0.830344	0.166873	0.0119676	0.38869	0.112336	0.675906	0.554439	0.888163	0.881613	0.348705
4	0.0130943	0.702295	0.417558	0.824496	0.271467	0.293717	0.655974	0.78709	0.66586	0.592221
5	0.993136	0.262433	0.132242	0.850283	0.521659	0.187733	0.334311	0.859878	0.534288	0.40397
6	0.500378	0.22558	0.647151	0.315375	0.647435	0.307202	0.382302	0.370105	0.425099	0.642584
7	0.429387	0.960168	0.128584	0.644386	0.539359	0.230073	0.822344	0.187343	0.768617	0.834613
8	0.494793	0.0896164	0.0811292	0.32082	0.12624	0.50482	0.545125	0.556587	0.822893	0.1666
9	0.776773	0.896758	0.119333	0.687507	0.92007	0.929793	0.0857724	0.595789	0.850245	0.861515
10	0.918536	0.966107	0.274968	0.801825	0.697714	0.295892	0.733026	0.654443	0.474951	0.162668
11	0.682397	0.590592	0.269285	0.845926	0.733575	0.296264	0.716101	0.660762	0.210495	0.244092
12	0.692736	0.363891	0.196404	0.479814	0.0263628	0.857153	0.770087	0.184756	0.944424	0.294371
13	0.00104153	0.799787	0.322529	0.274998	0.334355	0.00502477	0.543046	0.0934528	0.652469	0.502395
⋮	⋮	⋮	⋮	⋮	⋮	⋮	⋮	⋮	⋮	⋮
999989	0.38893	0.488278	0.0385544	0.495639	0.241208	0.580943	0.478893	0.49156	0.0338644	0.474437
999990	0.273409	0.729579	0.672658	0.306717	0.141573	0.619027	0.934958	0.498666	0.992331	0.96703
999991	0.914382	0.977408	0.0466257	0.353187	0.721385	0.220179	0.39309	0.243458	0.474306	0.423111
999992	0.0925393	0.0492193	0.497661	0.242698	0.201455	0.650363	0.908122	0.428147	0.428148	0.210207
999993	0.764645	0.948247	0.104478	0.823475	0.451709	0.332793	0.736104	0.991225	0.81367	0.507504
999994	0.227822	0.368746	0.951459	0.568322	0.3037	0.487387	0.152092	0.00259459	0.699173	0.49836
999995	0.862975	0.145	0.519696	0.812599	0.65848	0.649833	0.44456	0.630177	0.792391	0.168679
999996	0.313742	0.269196	0.0392782	0.90347	0.961205	0.743785	0.265006	0.513281	0.494936	0.768261
999997	0.347113	0.51485	0.293376	0.885858	0.947911	0.697525	0.369732	0.535832	0.211851	0.660482
999998	0.357896	0.573182	0.00617545	0.632811	0.579779	0.960782	0.888766	0.496995	0.384513	0.180331
999999	0.355143	0.69701	0.755443	0.18112	0.732943	0.569134	0.849911	0.443431	0.532987	0.716891
1000000	0.509224	0.903955	0.359278	0.178076	0.385685	0.0988106	0.885855	0.360228	0.446265	0.77072

@time map(sum, eachrow(df2));

  2.454178 seconds (60.12 M allocations: 1.056 GiB, 17.63% gc time, 4.70% compilation time)

@time map(sum, eachrow(df2));

  1.833568 seconds (59.99 M allocations: 1.050 GiB, 6.03% gc time)

@time map(sum, Tables.namedtupleiterator(df2));

  0.303801 seconds (1.22 M allocations: 66.451 MiB, 98.32% compilation time)

@time map(sum, Tables.namedtupleiterator(df2));

  0.004991 seconds (21 allocations: 7.631 MiB)

as you can see - this time it is much faster to iterate a type stable container still you might want to use the select syntax, which is optimized for such reductions:

this includes compilation time

@time select(df2, AsTable(:) => ByRow(sum) => "sum").sum

  0.483571 seconds (652.93 k allocations: 40.217 MiB, 2.73% gc time, 95.90% compilation time: 95% of which was recompilation)

1000000-element Vector{Float64}:
 4.582538267133092
 5.86712127030766
 4.85903797859344
 5.223773013475111
 5.07993464728864
 4.4632116807439814
 5.544873933254496
 3.7086242892505434
 6.723556847831017
 5.98012950282477
 ⋮
 3.70856059429051
 6.473850788533407
 4.2596556715720215
 5.684389196675713
 5.272159975659073
 5.46453064264186
 5.061230203019331
 5.834012167127027
 4.898097576410764

Do it again

@time select(df2, AsTable(:) => ByRow(sum) => "sum").sum

  0.006329 seconds (123 allocations: 7.634 MiB)

1000000-element Vector{Float64}:
 4.582538267133092
 5.86712127030766
 4.85903797859344
 5.223773013475111
 5.07993464728864
 4.4632116807439814
 5.544873933254496
 3.7086242892505434
 6.723556847831017
 5.98012950282477
 ⋮
 3.70856059429051
 6.473850788533407
 4.2596556715720215
 5.684389196675713
 5.272159975659073
 5.46453064264186
 5.061230203019331
 5.834012167127027
 4.898097576410764

---

This notebook was generated using Literate.jl.