vignettes: timeseries - ordered observations

[jangorecki]

There is wide range of features that data.table has in context of ordered observations, if one works with timeseries data it is definitely worth to be aware of those:

key - clustered index

rleid
shift
rolling funs (yet to be completed)
nafill, setnafill

between
inrange
rolling join
non-equi .EACHI join
foverlaps

IDate, ITime, IDateTime

hour, isoweek, mday, minute, month, quarter, second, wday, week, yday, year

as.data.table.xts, as.xts.data.table

nanotime pkg support in fwrite

[Henrik-P]

Hi @jangorecki!

A while ago I gave a data.table presentation at work, and now I realize that some of the material in the .Rmd file for the slides possibly may be of some use during the preparation of this vignette.
So far I have short text snippets, and in most cases visualizations, for the following topics:

• brief on normal join
• rolling join: nearest, Inf. -Inf, fixed distance
• non-equi join
• between & inrange
• rleid
• shift
• nafill & setnafill

Please let me know if this may be something you wish to have a look at. I'm happy to share. And of course: just feel free to ignore it.

[jangorecki]

@Henrik-P sure!

[Henrik-P]

OK, thanks! I'm not a very skilled github user, so I will need to re-read the Contribution guidelines on how to share the .Rmd file the proper way ;)

[Henrik-P]

@jangorecki A first draft of the .Rmd file is soon ready. Two questions of a more PR technical character: (1) Is it correct to assume that a test is not relevant for a vignette? At least I could find one for e.g. New vignette -- Usages of .SD. (2) Should I create a NEWS entry despite this is, at least from my perspective, very much work in progress?

Cheers

[jangorecki]

1. Yes, we don't need new unit tests for vignette. If you want to make your vignette future proof, one good option is to add hidden R chunks (include=F) in Rmd. In those chunks stopifnot()'s, so whenever a code will not meet expectations defined there, the vignette will terminate and not be built, we will easily spot that. But this is not really that important.
2. We will need NEWS entry ultimately, but vignette is likely to be evolving before it will be merged, so not really important yet.

Please ensure you push to Rdatatable org repo, so it will be easy for others to push commits there.

[MichaelChirico]

Yes, we don't need new unit tests for vignette. If you want to make your vignette future proof, one good option is to add hidden R chunks (include=F) in Rmd. In those chunks stopifnot()'s, so whenever a code will not meet expectations defined there, the vignette will terminate and not be built, we will easily spot that. But this is not really that important.

very nice trick!