Implement `reframe()`

[domq]

This is a successor to #102 of sorts.

• What I am trying to do:

  library(dplyr)
  library(multidplyr)
  # install.packages("broom")
  
  cluster <- new_cluster(2)
  iris %>% group_by(Species) %>% partition(cluster) %>% reframe(broom::tidy(lm(Petal.Length ~ Sepal.Length)))

• What I expect: the same data frame that I obtain when doing that without the %>% partition(cluster) part (modulo performance).
• What I get instead:

  Error in UseMethod("reframe") :
  no applicable method for 'reframe' applied to an object of class "multidplyr_party_df"
  

[hadley]

Thanks for the suggestion! Will definitely consider it when I'm next working on multidplyr.

[RudolfCardinal]

Dear @hadley , Hope all's well! I'm trying to shift from group_modify() to summarise() as you suggest in #102, but can't see a way to parallelize this within tidyverse. Apologies if I have missed it... I'm using dplyr 1.1.3, but I think 1.1.4 (current) is equivalent in these respects. First, evolution:

• I understand from your link that summarise() could then return an arbitrary rectangle. But, returning >1 row from summarise() gives a warning as being deprecated in dplyr 1.1.0 onwards. It suggests reframe() instead; fine.

But then, data provision:

• Is there any way to use reframe() to call a function with the subgroup data frame in its entirety? Or the "grouped-by" (key) and "other" (data) sections? Specifically, group_modify() provides the .y (key) and .x (subgroup data) pronouns (if that's the correct term) but I can't see a way to do it here.

For context, broadly what I am trying to do is:

• group_by();
• pass the key (grouped-by), which is a 1 * c1 tibble, and the data, an r * c2 tibble (or alternatively the whole chunk together, an r * [c1 + c2] tibble), to a function;
• have that function return a tibble with entirely different dimensions (a consistent number of columns, of course, but a variable number of rows);
• and parallelize that operation, since it should be an "embarrassingly parallel" problem.

Is there any way to achieve this? Thank you!

all the best,
Rudolf.