rand(::GroupedDataFrame) sampler?

[anandijain]

Hello,
I was wondering if there is a reason that a Sampler for the rand function is not practical for the GroupedDataFrame type.
It seems like a relatively easy and useful feature.
Thanks

[bkamins]

GroupedDataFrame does not allow duplicates and rand does return them in general if passed a dimension. Also it is easy enough to write gdf[rand(1:lengh(gdf))] if you want to pick a single group at random or [gdf[rand(1:lengh(gdf))] for _ in 1:how_many] if you want many of them.

[anandijain]

definitely, hear you on gdf[rand(1:lengh(gdf))]
I still don't fully understand why rand(gdf) shouldn't work though.

[bkamins]

rand(gdf) could work, but then we would not be able to support the whole rand API, which is defined as:

Pick a random element or array of random elements from the set of values specified by S

However it would be possible to support rand(gdf) and that is why I have not closed this issue to let people comment (this is a similar decision to: #2048, it is easy to add it, but it is also easy to have the functionality without adding it).

[bkamins]

After the discussion in #3437 I am closing this issue.