Skip to content

Latest commit

 

History

History
47 lines (24 loc) · 2.58 KB

TODO.md

File metadata and controls

47 lines (24 loc) · 2.58 KB
Raw

data.io To Do list

  • Should we always convert ts and mts objects into data frames? + see co2, nottem, pastecs::marbio, pastecs::marphy, pastecs::releve.

  • Also see tidyr::who, tidyr::population, vegan::BCI, vegan::mite, vegan::varespec, UsingR::babies, ca::wg93, faraway::eggs, mlbench::PimaIndianDiabetes2

  • Reimplement options(data.io.lang = "xx") as options(SciViews.lang = "xx"), and make changes in all SciViews packages (this option is not solely for {data.io} indeed).

  • Check the {units} package and verify there is no conflicts any more, or better, use it. Make sure one could use units(x) ans units(x) <- value or another form on all vectors.

  • Implement .DollarNames() to get a list of read() and write() types available, with a mechanism to add more types from additional packages (completion in R and RStudio for read$ and write$).

  • An easier form to indicate labels and units

  • Allow for data transformation and store info in a 'transfo' attribute. For instance, transforming "size [mm]" using log1p() with something like transfo(data = df, logsize = log1p(size)) would produce a transfo = "log1p" attribute in logsize. And when label is used in plots, we would get "log(size [mm] + 1)". Also a backtransfo() function to back transform variables.

  • dictionary() function that constructs a data dictionary for a labelled dataset (we probably need more attributes too). Would contain something like: name, label, units, transformation, type, dimensions, missing, examples, comment.

  • Prefer data.table::fread() and data.table::fwrite() for CSV files.

  • With {readr}, do not append attributes "spec" and "problems" to the object, but store them elsewhere (where?)

  • Use 'fst' package + 'qs' + 'vroom'.

  • Look for a support of 'datapackage.r'.

  • reimport print.subsettable() from {svMisc}

  • label versus comment for a dataframe?

  • Rework and modularize the read() function.

  • rio::characterize() or factorize() for SPSS or SAS datasets + more {rio}.

  • Add more recognized classes ('Date', 'time', etc.) for headers (+ abbrev?)

  • make iris.csvy, iris.csvy.zip, iris.sas7bdat and iris.syd working: add 'foreign' functions & 'csvy'.

  • Functions to write headers too: write() -> read() should restore same object!

  • Integrate {units} with the {quantities}, {convertr}, {prettyunits} and {units} packages.

  • A smaller example for read(), and detail all the cases in a vignette instead.

  • More tests!

  • {revdep}, see for instance, https://github.com/hadley/nycflights13