Insert dummy 0 value rows into dataframe

General Usage
1

Hello All,

I’m trying to figure out a quick way to insert dummy rows with 0 values if they are missing for a set of index fields

For example, let’s say I have this dataframe:


data = DataFrame(:index => String[], :metric => String[], :values => Int64[])

push!(data, ["a", "x", 1])
push!(data, ["a", "y", 2])
push!(data, ["b", "x", 1])


julia> data
3×3 DataFrame
 Row │ index   metric  values 
     │ String  String  Int64  
─────┼────────────────────────
   1 │ a       x            1
   2 │ a       y            2
   3 │ b       x            1

For this table, I have metrics x,y for index A, but only metric X for index B. Is there a simple way to populate this table with index b, metric y, value = 0?

Of course, in my real table, there are thousands/millions of rows, so I don’t want to loop through these values one at a time to insert them. In my current program, I pivot the data wide, insert 0 values where they are missing and then depivot it back, but not sure if there is an easy way to do that/better way to do that in Julia DataFrames.

Thanks.

2

There is a keyword argument cols in push! where you can do cols = :union do add new columns.

However you can’t control what the value will be, so you will need to do coalesce after your loop.

3

Sorry I’m not quite sure what you mean by this. How can I use union in this case?

4

What you want is https://github.com/JuliaData/DataFrames.jl/pull/1864 but it was never finished. If you would find this functionality useful I can re-write this PR anew and add it.

5

It does sound like what I want. I’d use the function if it existed.

Thanks

6

For now you can have a look at the internals of the implementation - they should be relatively fast (but I do not guarantee it as it was written a long time ago).

7

Please read the docstring for push! with ?push. Here is an MWE that points you in the right direction.

julia> data = DataFrame(:index => String[], :metric => String[], :values => Int64[])
0×3 DataFrame

julia> datas = [(index = "a", metric = "x"), (index = "a", values = 1)];

julia> for d in datas
           push!(data, d; cols = :union)
       end

julia> data
2×3 DataFrame
 Row │ index   metric   values  
     │ String  String?  Int64?  
─────┼──────────────────────────
   1 │ a       x        missing 
   2 │ a       missing        1
8

My solution, (when columns are categorical unique should be changed to levels)

d = DataFrame([(index = i, metric =  m) 
                            for i in unique(data.index)
                            for m in unique(data.metric)])

leftjoin(d, data, on =[:index, :metric])