Use case for mutate(clean = TRUE)

[ijlyttle]

Per Hadley's suggestion, this was moved from the manipulatr google group. link

Let's say I get a csv file with a bunch of variables named "XPF102", "FGR24D", and so on. There may be twenty of these.

From my data-dictionary, I see that "XPF102" is pressure in psi, "FGR24D" is a part-per-million of contaminant, and so on.

I have used plyr::summarise() to do two things at once (this could be what you are trying to get away from in dplyr):

data_new <- plyr::summarise(
  data_old,
  pressure = XPF102 * 6894.75729,  # convert to Pa from psi (I know a function is more appropriate here)
  concentration_contaminant = FGR24D / 1.e6, # convert to proportion from parts-per-million
  ...
)

Using dplyr, I might have to do this:

data_new <- 
  data_old %.%
  mutate(
    pressure = XPF102 * 6894.75729,
    concentration_contaminant = FGR24D / 1.e6, 
    ...
  ) %.%
  select(
    pressure,
    concentration_contaminant,
    ...
  )

Doing this, I seem to have the opportunity to mistype a variable name by violating DRY.

Hadley suggested:

Maybe an option to mutate like clean = TRUE?

I like the idea. My only (relatively uneducated) concern is if the user wants to name a variable clean, but I'm sure there is a clever way to avoid that.

Thanks,

Ian

[hadley]

I'm now slightly leaning towards a new verb called transmute().

[ijlyttle]

FWIW, I like it.

[piccolbo]

When I picked transmute for plyrmr I thought it was odd enough that there would be no sharing of it. I have no problem with transmute popping up in dplyr, quite the opposite, the only problem is that transmute in plyrmr is uber-general and allows you to do thing like multi-row summaries (e.g. quantiles) or expansions, like splitting a line of text into words. It evaluates the ... arguments in an expanded environment and binds them together in a data frame (vectors and data.frames and lists are all allowed), applies fractional recycling like cbind does and returns the result. The right name for this was probably transform but that was taken for a much more constrained operation. So I went with transmute, but now you need that name. Fine. So what do you suggest? I am willing to pick anything that will suggest absolute freedom in assembling the result, is in the dictionary and won't appear in dplyr in the next century.

[hadley]

It might be ok - what does the signature of plyrmr::transmute look like? We might be able to share the same generic.

[piccolbo]

function(.data, ..., .cbind = FALSE, .columns = NULL, .envir = parent.frame())

The problem I think is more the semantic difference. I know we had this discussion before, but it's not settled. Use case

mtcars %>% transmute(quantile(mpg), quantile(hp))