Feature request: Group data frames to accept string or symbol as index when grouping is by a single attribute

[alex-s-gardner]

#Make example grouped data frame

using DataFrames
df=DataFrame(city=["Paris", "London", "Paris", "Berlin", "London", "Berlin", "Berlin"],  
             date= ["10-1$k-2021"  for k in 3:9],
             v=38 .+100*rand(7))
gdf = DataFrames.groupby(df, :city)

this is how the data frame needs to be accessed now:

gdf[("Berlin", )]

for a data frame grouped by a single attribute it would be more intuitive to simply index in the same was as a data frame column

gdf["Berlin"] == gdf[:Berlin] ==  gdf[("Berlin", )]

[bkamins]

The reason why this is not allowed is that gdf[1] would be ambiguous as it could mean:

1. Selecting the first group.
2. Selecting the group with key equal to 1.

In the past we discussed allowing gdf(1) (and by extension e.g. gdf(1,2,3) for multiple grouping keys) to make this case easier, but it did not get much support. But maybe we can reconsider it.

CC @nalimilan @pdeffebach @kdpsingh

[alex-s-gardner]

Ahh.. I can see the challenge. DimensionalData.jl uses At(1)... such that gdf[1] is the first group and gdf[At(1)] is the group with key value of 1... though I'm not sure how simpatico that is with DataFrames

[bkamins]

We could use At (or other such wrapper), but in this case it is longer to write At(1) than (1,). And this was a consideration why we did not introduce it yet.

[alex-s-gardner]

At(1) is longer but more intuitive than (1,), but I certainly see your point. gdf(1) seems like a good compromise

[alex-s-gardner]

Thinking about this more gdf[At("Berlin")] produces more readable/understandable code as it retains the indexing brackets [] and then specifies what is being requested. gdf("Berlin") might be a bit ambiguous as it suggests a function call.