translation vignette feedback

[franknarf1]

Some thoughts re Rdatatable/data.table#3641 (comment) :

• filter is mistaken, I think, since DT[.(pred1, pred2)] attempts a join. This could be resolved by translating filter(pred1, pred2) to DT[pred1 & pred2] (maybe optimized by auto-indexing Extend auto indexing to multiple columns Rdatatable/data.table#1453) or DT[pred1][pred2]
• I think it's the case that grouped filter is generally faster using an alternative to the .SD way, though it is clunky: https://stackoverflow.com/a/16325932 On the other hand, improvements are planned for the .SD way: Further optimisation of .SD in j Rdatatable/data.table#735
• I did not test, but "mutate() allows to refer to variables that you just created" also applies to summarise(), which may also need a fix (eg mtcars %>% group_by(cyl) %>% summarise(n = n(), nfrac = n/nrow(.)))

[hadley]

Thanks!

I've fixed the translation for filter() so that it now reduces with &. Generate multiple calls [] feels risky to me — it will help in the case of multiple indexable expressions, but I suspect might otherwise slow down. For now, I'll leave that optimisation up to data.table.

Yes, I'm pinning my hopes on .SD being made faster. (I also asked a similar question at a similar time! https://stackoverflow.com/questions/16573995)

I'm going to leave the summarise() for now, since it's relatively uncommon in the wild, and I couldn't figure out how to handle it in dbplyr.

[hadley]

Added #75 to at least give the same error that dbplyr does for summarise() that refers to newly created errors.