How to read record format json in Julia? - julia

Question

I am able to read a json file and convert into dataframe using below code.
df = open(jsontable, "normal.json") |> DataFrame
normal.json looks like below,
{"col1":["thasin", "hello", "world"],"col2":[1,2,3],"col3":["abc", "def", "ghi"]}
So final df has,
3×3 DataFrame
│ Row │ col1 │ col2 │ col3 │
│ │ String │ Int64 │ String │
├─────┼────────┼───────┼────────┤
│ 1 │ thasin │ 1 │ abc │
│ 2 │ hello │ 2 │ def │
│ 3 │ world │ 3 │ ghi │
But, the same code is not working for record formatted json file.
the format is list like {column -> value}, … , {column -> value}
My sample json
{"billing_account_id":"0139A","credits":[],"invoice":{"month":"202003"},"cost_type":"regular"}
{"billing_account_id":"0139A","credits":[1.45],"invoice":{"month":"202003"},"cost_type":"regular"}
{"billing_account_id":"0139A","credits":[2.00, 3.56],"invoice":{"month":"202003"},"cost_type":"regular"}
Expected output:
billing_account_id cost_type credits invoice
0 0139A regular [] {'month': '202003'}
1 0139A regular [1.45] {'month': '202003'}
2 0139A regular [2.0, 3.56] {'month': '202003'}
This can be done in python like below,
data = []
for line in open("sample.json", 'r'):
data.append(json.loads(line))
print(data)
df=pd.DataFrame(data)
How to do this in Julia?

Answer 1

Note that your file is not a valid JSON (its lines are valid JSON, not the whole file).
You can do this like this:
julia> using DataFrames, JSON3
julia> df = JSON3.read.(eachline("sample.json")) |> DataFrame;
julia> df.credits = Vector{Float64}.(df.credits);
julia> df.invoice = Dict{Symbol,String}.(df.invoice);
julia> df
3×4 DataFrame
│ Row │ billing_account_id │ credits │ invoice │ cost_type │
│ │ String │ Array{Float64,1} │ Dict{Symbol,String} │ String │
├─────┼────────────────────┼────────────────────────────┼────────────────────────┼───────────┤
│ 1 │ 0139A │ 0-element Array{Float64,1} │ Dict(:month=>"202003") │ regular │
│ 2 │ 0139A │ [1.45] │ Dict(:month=>"202003") │ regular │
│ 3 │ 0139A │ [2.0, 3.56] │ Dict(:month=>"202003") │ regular │
The transformations on :credits and :invoice columns are to make them of type that is easy to work with (otherwise they use types that are defined internally by JSON3.jl).
A more advanced option is to do it in one shot by specifying the row schema using a NamedTuple type e.g.:
julia> df = JSON3.read.(eachline("sample.json"),
NamedTuple{(:billing_account_id, :credits, :invoice, :cost_type),Tuple{String,Vector{Float64},Dict{String,String},String}}) |>
DataFrame
3×4 DataFrame
│ Row │ billing_account_id │ credits │ invoice │ cost_type │
│ │ String │ Array{Float64,1} │ Dict{String,String} │ String │
├─────┼────────────────────┼────────────────────────────┼─────────────────────────┼───────────┤
│ 1 │ 0139A │ 0-element Array{Float64,1} │ Dict("month"=>"202003") │ regular │
│ 2 │ 0139A │ [1.45] │ Dict("month"=>"202003") │ regular │
│ 3 │ 0139A │ [2.0, 3.56] │ Dict("month"=>"202003") │ regular │

Answer 2

Unrelated to the julia answer, but in python you can do
pd.read_json("sample.json", lines=True)