filter and broadcasting for GroupedDataFrame

[bkamins]

This can be done post 1.0, but I think we are ready to decide that GroupedDataFrame is a collection of SubDataFrames, so:

1. we could add filter to it (question just asked on Slack)
2. we could consider enabling broadcasting of GroupedDataFrame - here the only decision would be if the result of such broadcasting operation should be a Vector or a GroupedDataFrame (both make sense in different contexts; I tend to prefer a Vector as map already gives an option with GroupedDataFrame).

[akdor1154]

just ran into the need to filter groups on my first foray into Julia. It actually looks impossible to hack together something myself (based on my three days' experience with the language, so I'm probably wrong here) so it would be nice to see this included as a function.

[bkamins]

It is not that hard - just use Bool indexing into GroupedDataFrame:

julia> using Statistics

julia> df = DataFrame(g = 1:10, x = rand(10))
10×2 DataFrame
│ Row │ g     │ x         │
│     │ Int64 │ Float64   │
├─────┼───────┼───────────┤
│ 1   │ 1     │ 0.109814  │
│ 2   │ 2     │ 0.966174  │
│ 3   │ 3     │ 0.0975521 │
│ 4   │ 4     │ 0.393241  │
│ 5   │ 5     │ 0.0790575 │
│ 6   │ 6     │ 0.809334  │
│ 7   │ 7     │ 0.756541  │
│ 8   │ 8     │ 0.735663  │
│ 9   │ 9     │ 0.0790629 │
│ 10  │ 10    │ 0.244876  │

julia> gdf = groupby(df, :g)
GroupedDataFrame with 10 groups based on key: g
First Group (1 row): g = 1
│ Row │ g     │ x        │
│     │ Int64 │ Float64  │
├─────┼───────┼──────────┤
│ 1   │ 1     │ 0.109814 │
⋮
Last Group (1 row): g = 10
│ Row │ g     │ x        │
│     │ Int64 │ Float64  │
├─────┼───────┼──────────┤
│ 1   │ 10    │ 0.244876 │

julia> gdf[[mean(sdf.x) > 0.5 for  sdf in gdf]]
GroupedDataFrame with 4 groups based on key: g
First Group (1 row): g = 2
│ Row │ g     │ x        │
│     │ Int64 │ Float64  │
├─────┼───────┼──────────┤
│ 1   │ 2     │ 0.966174 │
⋮
Last Group (1 row): g = 8
│ Row │ g     │ x        │
│     │ Int64 │ Float64  │
├─────┼───────┼──────────┤

and the result is a GroupedDataFrame containing only groups for which mean of :x column is greater than 0.5. Still probably this:

filter(sdf -> mean(sdf.x) > 0.5, gdf)

or (this syntax will be available soon and will be faster):

filter(:x => x -> mean(x) > 0.5, gdf)

is more convenient.

[akdor1154]

ah, thanks for the workaround! no offence taken if you want to clean up my comments to this issue to keep it clean :)

[bkamins]

Actually thank you for commenting - we need end user feedback.