DataFrames and DataTables support

[nalimilan]

In #145 RCall was changed to create NullableArray and CategoricalArray objects rather than DataArray and PooledDataArray. This was consistent with the DataFrames plans at the time.

This decision probably needs revisiting now that the plans have changed (i.e. DataFrames will keep using DataArray and PooledDataArray, and the new DataTables package uses NullableArray and CategoricalArray). The current state is inconsistent since DataFrame is not made to work well with NullableArray and CategoricalArray. So there are two options:

• keep creating DataFrame objects by default, and go back to creating DataArray and PooledDataArray columns
• switch to DataTable and keep creating NullableArray and CategoricalArray columns

I'm really torn about which solution is better. We don't have all the high-level query APIs yet to make working with DataTables as nice as it should be. Yet, going back to DataArray could be annoying for users who already had to adapt their code. Maybe we should do it nevertheless until the Nullable API stabilizes, to avoid having our data frame/table ecosystem too inconsistent or painful to use for several months.

[dmbates]

@nalimilan I support the idea of going back to DataArrays and PooledDataArrays in RCall for the short term. I think it would be much less confusing to make the switch to CategoricalArrays, NullableArrays, DataTables and StatsModels in one step.

When I am teaching courses for to R users by far the simplest approach is to use R to load the data and then use RCall to transfer.

@randy3k, @simonbyrne Let me know if I can help, should you decide that this is a good path forward.

[randy3k]

I also agree that we should switch back to DataArrays. In fact, I am working on the idea that I am @simonbyrne discussed earlier at #138 (comment). Once it is done, there are converters like, convert(DataFrame, foobar) and convert(DataTable, foobar) and the default converter rcopy. I am thinking also some kind of register to allow users to choose their preferred default converter.

[nalimilan]

OK, let's go back to DataArrays then.