Summary table of unique value combinations in DataFrames.jl

Question

I often want to find the unique combinations of some grouping variables in a data table. With R + dplyr, my normal workflow is to combine groupby(data, var1, var2, var3) %>% summarise, which returns a new table with the columns var1, var2, var3, with one row for each unique combination of values found in data.

What's the idiomatic way to do this in DataFrames.jl?

Answer 1

In DataFrames.jl, a DataFrame is a collection of rows. So the right mental model here is to first select only the columns you care about, then get the unique rows from that table, as in

select(data, [:var1, :var2, :var3]) |> unique!

(Or if you hate the pipe/love extra parens:

unique!(select(data, [:var1, :var2, :var3]))

unique! is recommended here because select makes a copy of the underlying columns. Alternatively, you could use a view or indexing, but these require unique (which does not mutate the underlying column vectors) so as not to corrupt the original data frame:

unique(data[!, [:var1, :var2, :var3]])
unique(view(data, :, [:var1, :var2, :var3]))

Answer 2

Alternatively you can write:

keys(groupby(data, [:var1, :var2, :var3]))

to get a vector of unique grouping keys. Then you can collect them to a DataFrame if you want by writing:

groupby(data, [:var1, :var2, :var3]) |> keys |> DataFrame