group function

[matthieugomez]

The group function allows to combine multiple columns of a dataframe into one PooledDataArray column. A user case is to create groups based on multiple variables before fitting a model.
The function group basically rewraps code from groupby.I'm not sure whether this functionality already exists or not.

[tshort]

Needs tests and docs...

[matthieugomez]

I have added tests & a doc. I have added a skipna option : should observations with some column equal to NA grouped into a NA group? The option defaults to true. This returns NA for any observation with a original column equals to NA, which is generally what the user wants (especially when using poolall before fitting models). I don't know the usual name for NA options in Julia, so tell me if skipna does not sound right.

[matthieugomez]

I have updated my commit to solve a bug I spotted in groupby, which gives incorrect groups in the following case:

   df = DataFrame(v1 = pool(1:1000), v2 = pool(fill(1, 1000))) 
   groupby(df, [:v1, :v2])

Grouping by v1 and v2 should give 1000 groups, but only 255 groups are created. This happens because the type of refs element for the last column of df is Uint8, and this type is not promoted when grouping by v1 and v2, even though the number of groups becomes > 255. I now promote this vector to Uint32 in all cases for intermediary computations & I have added a test to catch this case.

[matthieugomez]

I think it's good on my end now.