[R] Can I pass the grouped portions of a dataframe/tibble to a function in dplyr

[Chris Evans]

Ouch.  I should have know all those points Rui: my bad.  Casual behaviour while just rushing up a little example. Good to be reminded. 

group_modify()  is clearly exactly what I wanted and I will experiment with it and make sure I understand it properly.  I see from the help that it evolves from, or supercedes aspects of do() which I think must have been the function I had forgotten.  Even more interestingly I see that it seems to lead me into interesting options and experimental developments in tidyverse that I didn't know.

Excellent.  Perfect help ... many thanks!

Chris

----- Original Message -----
> From: "Rui Barradas" <ruipbarradas using sapo.pt>
> To: "Chris Evans" <chrishold using psyctc.org>, "R-help" <r-help using r-project.org>
> Sent: Sunday, 5 July, 2020 13:16:19
> Subject: Re: [R] Can I pass the grouped portions of a dataframe/tibble to a function in dplyr

> Hello,
> 
> I forgot to say I redid the data set setting the RNG seed first.
> 
> 
> 
> set.seed(2020)
> n <- 50
> x <- 1:n
> y <- sample(1:3, n, replace = TRUE)
> z <- rnorm(n)
> tib <- tibble(x,y,z)
> 
> 
> Also, don't do
> 
> as_tibble(cbind(...))
> as.data.frame(cbind(...))
> 
> 
> If one of the variables is of a different class (example, "character")
> all variables are coerced to the least common denominator. It's much
> better to call tibble() or data.frame() directly.
> 
> Hope this helps,
> 
> Rui Barradas
> 
> 
> Às 12:04 de 05/07/2020, Rui Barradas escreveu:
>> Hello,
>> 
>> You can pass a grouped tibble to a function with grouped_modify but the
>> function must return a data.frame (or similar).
>> 
>> ## this will also do it
>> #sillyFun <- function(tib){
>> #  tibble(nrow = nrow(tib), ncol = ncol(tib))
>> #}
>> 
>> 
>> sillyFun <- function(tib){
>>    data.frame(nrow = nrow(tib), ncol = ncol(tib)))
>> }
>> 
>> tib %>%
>>    group_by(y) %>%
>>    group_modify(~ sillyFun(.))
>> ## A tibble: 3 x 3
>> ## Groups:   y [3]
>> #      y  nrow  ncol
>> #  <dbl> <int> <int>
>> #1     1    17     2
>> #2     2    21     2
>> #3     3    12     2
>> 
>> 
>> Hope this helps,
>> 
>> Rui Barradas
>> 
>> Às 09:43 de 05/07/2020, Chris Evans escreveu:
>>> Apologies if this is a stupid question but searching keeps getting
>>> things I know and don't need.
>>>
>>> What I want to do is to use the group-by() power of dplyr to run
>>> functions that expect a dataframe/tibble per group but I can't see how
>>> do it. Here is a reproducible example.
>>>
>>> ### create trivial tibble
>>> n <- 50
>>> x <- 1:n
>>> y <- sample(1:3, n, replace = TRUE)
>>> z <- rnorm(n)
>>> tib <- as_tibble(cbind(x,y,z))
>>>
>>> ### create trivial function that expects a tibble/data frame
>>> sillyFun <- function(tib){
>>> return(list(nrow = nrow(tib),
>>> ncol = ncol(tib)))
>>> }
>>>
>>> ### works fine on the whole tibble
>>> tib %>%
>>> summarise(dim = list(sillyFun(.))) %>%
>>> unnest_wider(dim)
>>>
>>> That gives me:
>>> # A tibble: 1 x 2
>>>     nrow  ncol
>>>    <int> <int>
>>> 1    50     3
>>>
>>>
>>> ### So I try the following hoping to apply the function to the grouped
>>> tibble
>>> tib %>%
>>> group_by(y) %>%
>>> summarise(dim = list(sillyFun(.))) %>%
>>> unnest_wider(dim)
>>>
>>> ### But that gives me:
>>> # A tibble: 3 x 3
>>>        y  nrow  ncol
>>>    <dbl> <int> <int>
>>> 1     1    50     3
>>> 2     2    50     3
>>> 3     3    50     3
>>>
>>> Clearly "." is still passing the whole tibble, not the grouped
>>> subsets.  What I can't find is whether there is an alternative to "."
>>> that would pass just the grouped subset of the tibble.
>>>
>>> I have bodged my way around this by writing a function that takes
>>> individual columns and reassembles them into a data frame that the
>>> actual functions I need to use require but that takes me back to a lot
>>> of clumsiness both selecting the variables to pass in the dplyr call
>>> to the function and putting the reassemble-to-data-frame bit in the
>>> function I call.  (The functions I really need are reliability
>>> explorations and can called on whole dataframes.)
>>>
>>> I know I can do this using base R split and lapply but I feel sure it
>>> must be possible to do this within dplyr/tidyverse.  I'm slowly
>>> transferring most of my code to the tidyverse and hitting frustrations
>>> but also finding that it does really help me program more sensibly,
>>> handle relational data structures more easily, and write code that I
>>> seem better at reading when I come back to it after months on other
>>> things so I am slowly trying to move all my coding to tidyverse.  If I
>>> could see how to do this, it would help.
>>>
>>> Very sorry if the answer should be blindingly obvious to me.  I'd also
>>> love to have pointers to guidance to the tidyverse written for people
>>> who aren't professional coders or statisticians and that go a bit
>>> beyond the obvious basics of tidyverse into issues like this.
>>>
>>> TIA,
>>>
>>> Chris
>>>
>> 
> 
> --
> Este e-mail foi verificado em termos de vírus pelo software antivírus Avast.
> https://www.avast.com/antivirus

-- 
Small contribution in our coronavirus rigours: 
https://www.coresystemtrust.org.uk/home/free-options-to-replace-paper-core-forms-during-the-coronavirus-pandemic/

Chris Evans <chris using psyctc.org> Visiting Professor, University of Sheffield <chris.evans using sheffield.ac.uk>
I do some consultation work for the University of Roehampton <chris.evans using roehampton.ac.uk> and other places
but <chris using psyctc.org> remains my main Email address.  I have a work web site at:
   https://www.psyctc.org/psyctc/
and a site I manage for CORE and CORE system trust at:
   http://www.coresystemtrust.org.uk/
I have "semigrated" to France, see: 
   https://www.psyctc.org/pelerinage2016/semigrating-to-france/ 
   https://www.psyctc.org/pelerinage2016/register-to-get-updates-from-pelerinage2016/

If you want an Emeeting, I am trying to keep them to Thursdays and my diary is at:
   https://www.psyctc.org/pelerinage2016/ceworkdiary/
Beware: French time, generally an hour ahead of UK.