# Creating a 3d frequency array for categorical variables from a dataframe
**URL:** https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163
**Category:** New to Julia
**Tags:** dataframes
**Created:** [February 12, 2021, 5:39pm UTC](https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163 "2021-02-12T17:39:45Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@samerb](https://discourse.julialang.org/u/samerb)
#### Post date: [February 12, 2021, 5:39pm UTC](https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163/1 "2021-02-12T17:39:45Z")
I have a dataframe with three categorical variables `a,b,c`. taking values in 1:2, 1:5, and 1:7 respectively. Then I calculate the number rows with each of the possible 2\*5\*7=70 combinations of a,b,c:
```julia
df = transform(groupby(df, [:a,:b,:c]), nrow => :count)
df = unique(df[:, [:a, :b, :c, :count])
```
Now I want to create 3d 2\*5\*7 array called `frequency` such that frequency[i,j,k] is equal to the value of `count` in the row corresponding to a=i,b=j,c=k.
How can I do this? I suspect the `unstack` function may help but I don’t understand that function well.
---
### Author:  [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)
#### Post date: [February 12, 2021, 8:30pm UTC](https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163/2 "2021-02-12T20:30:05Z")
Here’s what I would do using DataFramesMeta and Chain.jl
```julia
julia> using Chain, DataFramesMeta;
julia> df = DataFrame(a = rand(1:2, N), b = rand(1:5, N), c = rand(1:7, N));
julia> df_count = @chain df begin
groupby([:a, :b, :c])
@combine(count = length(:a))
end;
julia> combinations = reshape(collect(Iterators.product(1:2, 1:5, 1:7)), :);
julia> df_full = @chain combinations begin
DataFrame
rename!([:a, :b, :c])
end;
julia> df_fraction = @chain df_full begin
leftjoin(df_count, on = [:a, :b, :c])
@transform(count = coalesce(:count, 0))
@transform(fraction = :count ./ sum(:count))
end;
```
---
### Author:  [@samerb](https://discourse.julialang.org/u/samerb)
#### Post date: [February 12, 2021, 9:56pm UTC](https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163/3 "2021-02-12T21:56:29Z")
Thanks but I’m not seeing how to access the frequency at particular indices. Suppose for I want to know the frequency with which `a=1,b=2,c=3` occurs. I guess I could convert the dataframe into a matrix and then loop over every row in the matrix and say `if matrix[_,1] = 1 && matrix[_,2]=2, && matrix[_,3] = 3`, then give me `matrix[_,4]`. But I figured there would be a more systematic way.
---
### Author:  [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)
#### Post date: [February 12, 2021, 10:20pm UTC](https://discourse.julialang.org/t/creating-a-3d-frequency-array-for-categorical-variables-from-a-dataframe/55163/4 "2021-02-12T22:20:31Z")
Ah yes I see now.
This is is a commonly requested feature in DataFrames. There is no way to index columns for an easy lookup like that.
You can make a `GroupedDataFrame` grouping on `[:a, :b, :c]` and then index like
```julia
gd[(a = 1, b = 2, c = 3)]
```
This will be fast since a `GroupedDataFrame` creates a hash for lookup just like a `Dict`.
However this will return a `SubDataFrame`, which is itself not the nicest object.
I would do the following
```julia
julia> gd = groupby(df_fraction, [:a, :b, :c]);
julia> function get_fraction(gd, ;a = nothing, b = nothing, c = nothing)
only(gd[(;a, b, c)]).fraction
end
get_fraction (generic function with 1 method)
julia> get_fraction(gd; a = 1, b = 2, c = 3)
0.018
```
```julia
```