sl-solution

For scenarios like `view(ds, :, :)` it is great, however, my problem is about the situation when a view has a non consecutive permutation of rows, this happens for `groupby`...

If that's the case, does the following help? ```julia julia> _tmp = Tables.getcolumn(sds, :x1) # this is expensive but we can do it once julia> _tmp[1] ```

but here _tmp isn't the whole column it is like a pointer to the actual values.

I see, let me think more about this.

> in Pandas: > > ```python > df = pd.DataFrame({ > 'paddockId': [0, 0, 1, 1, 2, 2], > 'color': ['red', 'blue', 'red', 'blue', 'red', 'blue'], > 'count': [3, 4,...

@bkamins that seems ok, I am currently working on its performance and keep updating it. Adding one more thing, I know in the current design, tabulating data (pivot and aggregation)...

Maybe it needs a little more thinking (BTW I think something like `table`, `tabulate`,... is a better name because it is really more than pivoting), few things * the new...

> For me this functionality is mostly for transforming data.... It is indeed just aggregation and transposing in some sense. And aggregation and the proposed df_transpose() can handle it already,...

> [On Slack](https://julialang.slack.com/archives/CQQ32Q9C5/p1599642920077900), I wrote that I like the new `tidyr::pivot_` functions (https://tidyr.tidyverse.org/articles/pivot.html) because the names make it really obvious what they do, and they have good arguments for indicating...

Just one thing to add is that the situation for median should be similar to `std` example, i.e. 4x faster, and in this case we can easily save seconds for...