A place for compose about magrittr HOT 18 CLOSED

+1 (via your Twitter feed). I'll have to think through the specifics more, but I see many use cases.

from magrittr.

Could this be used with with() to build a proper data = argument into functions that lack it? That would be a big win.

from magrittr.

I think that it will be very useful to write compact anonymous functions with intermediate computed quantities. Looks great.

from magrittr.

@jennybc

For this purpose I've been using this simple operator lately with good success.

`%within%` <- function(lhs, rhs) {
  lhs <- as.list(lhs)
  rhs <- substitute(rhs)
  eval(rhs, envir = lhs, enclos = parent.frame())
}

@smbache Do you think %within% could be magrittr material?

from magrittr.

Just a remark: %within% exists in lubridate, and so could cause conflicts
when people want to use lubridate in combination with dplyr (which
depends on magrittr, right?)

On Tue, Jun 17, 2014 at 3:57 PM, lionelgit [email protected] wrote:

@jennybc https://github.com/jennybc

For this purpose I've been using this simple operator lately with good
success.

%within% <- function(lhs, rhs) {
lhs <- as.list(lhs)
rhs <- substitute(rhs)
eval(rhs, envir = lhs, enclos = parent.frame())
}

@smbache https://github.com/smbache Do you think %within% could be
magrittr material?

—
Reply to this email directly or view it on GitHub
#18 (comment).

from magrittr.

@jennybc I am not sure exactly what you'd like. Could you give an example of how you imagine this should work?

from magrittr.

@lionelgit Can you give examples of some use cases?

from magrittr.

@jennybc although not relevant for your comment as such, it did make me think of the option to specify which arguments the composed function should accept in addition to the first one. For now, I added ellipses to the signature, so they can be used in the some of the nested functions:

library(dplyr)
library(magrittr)
f <- compose(
  x ~ x %>% group_by(Species), 
  y ~ y %>% do(arrange(., ...)), 
  z ~ z %>% do(head(., 2))
)

iris %>% f(Sepal.Length)
iris %>% f(Sepal.Width)

I might have a look at more general way of being able to specify the signature of the composed function.

from magrittr.

@holstius Indeed it's difficult to avoid name clashing with the hadleyverse... It's a shame that such an expressive operator is reserved for such a minor function though.

@jennybc @smbache

You use it simply like this:

data %within%
  lm(col1 ~ col2)

It will be as if you did

attach(data)
lm(col1 ~ col2)
detach(data)

from magrittr.

I have now made the way to add arguments to the composed function a little more flexible:

mae <- compose(abs, x ~ mean(x, na.rm = na.rm),
               .args = list(na.rm = TRUE))

The above function can now be written as

f <- compose(
  x ~ x %>% group_by(Species), 
  y ~ y %>% do(arrange(., ...)), 
  z ~ z %>% do(head(., 2)),
  .args = list(...)
)

I think this adds a lot of flexibility in composing the functions.

from magrittr.

@lionelgit Right, the name is a little reversed though, as lm is evaluated within the data, and not the other way around. Perhaps %expose% as it exposes the data for the rhs.

from magrittr.

I was thinking of it as putting the columns of data inside the next function, but you're right.

Maybe %wrap% ?

from magrittr.

yeah, perhaps :) naming is not always easy.

from magrittr.

I am not sure that an operator is really needed in magrittr for this, as one can just do:

data %>%
with(lm(col1, col2))

The operator does shorten it a bit...

from magrittr.

It's also more expressive. Well, up to you :)

from magrittr.

@smbache Here's a simple example of the sort of usage I mean. The cor() function lacks both a data = and subset = argument. So if I want to compute correlation of two variables in a data.frame for a subset of the rows, it's kind of awkward. Does magrittr already offer a way to do what I do below less awkwardly? Or is it relevant to this conversation about composition? Either way, I'll be happy!

gdURL <- "http://www.stat.ubc.ca/~jenny/notOcto/STAT545A/examples/gapminder/data/gapminderDataFiveYear.txt"
gDat <- read.delim(file = gdURL)
str(gDat) # 'data.frame':  1704 obs. of  6 variables:

with(subset(gDat, subset = country == "Colombia"), cor(lifeExp, gdpPercap))
# [1] 0.9514699

from magrittr.

@jennybc @lionelgit I added a version of the suggested wrap/expose operator under the name %$%, where the rationale of the name is that you get to not write data$ in front of the names. You could then do, say

iris %$% cor(Sepal.Length, Sepal.Width)

or

iris %>%
subset(Species == "setosa") %$%
cor(Sepal.Length, Sepal.Width)

As to existing options, in combination with dplyr, you could do

iris %>%
filter(Species == "setosa") %>%
summarise(Correlation = cor(Sepal.Length, Sepal.Width))

or use a lambda/anonymous function.

I am not sure whether it will be included in the next version, I also have to consider the opinion of my co-author of the package. But it is sort of handy, I admit. :)

from magrittr.

The %$% is a great idea. And for those using a functional editor, you can use this unicode character which conveys the idea of sending the cols in the function that follows :)

⇉

from magrittr.