Only sum and mean are benchmarked currently, albeit on a reasonable range of cardinalities and data types.
A ton quickly springs to mind to add, so it's more a question of balancing resources with potential insight.
Perhaps for starters we can consider the following.
Add median. It's a different ball game to sum and mean to calculate and is also very common in data science.
Add a very simple user defined function (see #57 too); e.g.
DT[, max(colB) - min(colB), by=colA]
vs
DT[, range(colB), by=colA].
Or something like DT[, 100*mean(colB), by=colA].
1 and 2 are still single aggregates though. Also common in data science is returning more than 1 value from each group, as follows.
largest 2 values in ColB by group
DT[, colB[head(order(ColB), 2)], by=colA]
and adding a column by group ( scale springs to mind ) :
jangorecki
commented
5 years ago
Only
sumandmeanare benchmarked currently, albeit on a reasonable range of cardinalities and data types. A ton quickly springs to mind to add, so it's more a question of balancing resources with potential insight. Perhaps for starters we can consider the following.1 and 2 are still single aggregates though. Also common in data science is returning more than 1 value from each group, as follows.
and adding a column by group ( scale springs to mind ) :
All of these new tests would need to be added for each product.