Stream: helpdesk (published)

Topic: ✔ Traversing Nested Dictionary

Davi Sales Barreira (Oct 18 2021 at 14:22):

Hey everyone, I was wondering if there was an efficient ("canonical") way to traverse a dictionary with a JSON like structure. For example, I have a dictionary with an array of dictionaries that sometimes don't have the same keys. I then need to iterate over these dictionaries and check if an specific key has a certain value. For example:

for cell in cells
  if cell["key1"]["key2"]["key3"] == true
     # do stuff
  end
end

The issue is that sometimes the cell won't have either "key1", or "key2" or "key3", so a simple get(cell, key1, false) won't work. Of course, I could always write a bunch of if statements, but I was wondering if there was a smarter and more direct way of doing this.

Richard Reeve (Oct 18 2021 at 15:40):

Well, it's hardly elegant, but you can chain gets by putting empty Dicts in instead of false:

if get(get(get(cell, "key1", Dict()), "key2", Dict()), "key3", nothing) == "value"
  # Do stuff
end

Or, perhaps marginally less inelegant:

if cell |> x -> get(x, "key1", Dict()) |> x -> get(x, "key2", Dict()) |> x -> get(x, "key3", nothing) == "value"
  # Do something
end

Davi Sales Barreira (Oct 18 2021 at 15:44):

Richard Reeve said:

Well, it's hardly elegant, but you can chain gets by putting empty Dicts in instead of false:

if get(get(get(cell, "key1", Dict()), "key2", Dict()), "key3", nothing) == "value"
  # Do stuff
end

Or, perhaps marginally less inelegant:

if cell |> x -> get(x, "key1", Dict()) |> x -> get(x, "key2", Dict()) |> x -> get(x, "key3", nothing) == "value"
  # Do something
end

Thanks, the empty Dict seems like a good idea. Putting it in a function and thus allowing an arbitrary number of keys looks like a nice work around.

Kwaku Oskin (Oct 18 2021 at 16:00):

You may write your own get function, which supports nested getting.

Kwaku Oskin (Oct 18 2021 at 16:01):

Is better then Dict(), since it will be less allocating and as such giving less pressure to gc.

Davi Sales Barreira (Oct 18 2021 at 16:17):

Can you give an example?

Davi Sales Barreira (Oct 18 2021 at 16:17):

I'm trying to write a get function that parses the nested dict, but I'm not finding a clever way of doing so.

Richard Reeve (Oct 18 2021 at 16:47):

julia> function nested_get(dict::Dict, keys::AbstractVector, default = nothing)
         if isempty(keys)
           return dict
         end
         key = popfirst!(keys)
         if haskey(dict, key)
           return nested_get(dict[key], keys)
         end
         return default
       end
nested_get (generic function with 2 methods)

julia> function nested_get(non_dict, keys::AbstractVector, default = nothing)
         if isempty(keys)
           return non_dict
         end
         return default
       end
nested_get (generic function with 4 methods)

julia> cells = Dict{String, Any}("a" => 1)
Dict{String, Any} with 1 entry:
  "a" => 1

julia> nested_get(cells, ["key1", "key2", "key3"]) === nothing
true

julia> cells["key1"]=Dict("key2" => Dict("key3"=>1))
Dict{String, Dict{String, Int64}} with 1 entry:
  "key2" => Dict("key3"=>1)

julia> nested_get(cells, ["key1", "key2", "key3"])
1

So it'll return nothing if there's no match at any point, and it'll return the value if there's a full match. Something like that?

Richard Reeve (Oct 18 2021 at 16:51):

It's more complicated than what I suggested originally, but Andrey is right that this is a better solution (assuming that's what he had in mind!).

Kwaku Oskin (Oct 18 2021 at 17:11):

Yes, that was the idea.
Instead of nothing you can provide default value, as in usual get function.

Kwaku Oskin (Oct 18 2021 at 17:11):

Oh!

Kwaku Oskin (Oct 18 2021 at 17:11):

You did :-)

Richard Reeve (Oct 18 2021 at 17:11):

Yes, I just updated it to do that!

Kwaku Oskin (Oct 18 2021 at 17:12):

You can use splatting instead of vector, but this is just a matter of taste.

Davi Sales Barreira (Oct 18 2021 at 18:20):

Thanks @Richard Reeve , it works beautifully. I actually prefer the vector of keys than the splatting. :)
It gives a sense that the order is important (which it is), while just splatting gives me the sense that they are somehow parallel.

Davi Sales Barreira (Oct 18 2021 at 18:21):

This function should be in Base, it's very neat, and I constantly need something like it.

Davi Sales Barreira (Oct 18 2021 at 18:22):

And it fits perfectly as another multiple dispatch for get.

Kwaku Oskin (Oct 18 2021 at 18:34):

If we do some type piracy, then vector gives funny notation
d[["key1", "key2", "key3"]], while splatting d["key1", "key2", "key3"]

Honestly, I do not know which one is better.

Notification Bot (Oct 19 2021 at 16:46):

Davi Sales Barreira has marked this topic as resolved.

Notification Bot (Oct 19 2021 at 16:46):

Davi Sales Barreira has marked this topic as unresolved.

Notification Bot (Oct 19 2021 at 16:46):

Davi Sales Barreira has marked this topic as resolved.

Last updated: Nov 06 2024 at 04:40 UTC