courtiol / lay Goto Github PK

I am not sure why but slider (and only slider) shows very different speed whenever run in README vs in the vignette showing the benchmarks:

Vignette:

README:

@DavisVaughan, any idea why? I only discovered your great package right now and did not dig, but this is quite intriguing... Is it a compilation issue?

It does not make much sense to do means and medians of different sepal dimensions.
We need a better example.
Use marketting data with multiple Q_r_ questions?

It would be great if lay() could allow the passing of arguments.
I cannot get it work with a pmap() (hence no PR; forwarding the lay-level ... seems more tricky than I thought) but here is an apply()-based example:

    library(dplyr, warn.conflicts = FALSE)

    lay <- function(x, fun, ...) {
      fn <- rlang::as_function(fun)
      bits <- apply(x, 1, fn, ...)
      vctrs::vec_c(!!!bits)
    }

    iris <- as_tibble(iris)
    iris_na <- iris
    iris_na[1, 1] <- NA

    iris_na %>%
      mutate(sepal = lay(across(starts_with("Sepal")), median)) %>%
      slice_head(n = 2)
    #> # A tibble: 2 x 6
    #>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species sepal
    #>          <dbl>       <dbl>        <dbl>       <dbl> <fct>   <dbl>
    #> 1         NA           3.5          1.4         0.2 setosa  NA   
    #> 2          4.9         3            1.4         0.2 setosa   3.95

    iris_na %>%
      mutate(sepal = lay(across(starts_with("Sepal")), median, na.rm = TRUE)) %>%
      slice_head(n = 2)
    #> # A tibble: 2 x 6
    #>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species sepal
    #>          <dbl>       <dbl>        <dbl>       <dbl> <fct>   <dbl>
    #> 1         NA           3.5          1.4         0.2 setosa   3.5 
    #> 2          4.9         3            1.4         0.2 setosa   3.95

We need to design a nice hexsticker... perhaps a hammock with a tibble sleeping in it?

"Benchmarks" and "Benchmark on smaller dataset" are not consistent...

I need to break down the examples in the README for visual purposes.

I need benchmark to compare the performances of multiple rowwise alternatives.

Users unfamiliar with lay may call rowwise() beforehand.
Doing so should exert no effect besides slowing down computation massively.
So perhaps we should detect rowwise groupings and act upon ithat (remove temporarily, warn, err, ...?).
That would probably imply to detect if the object is a tibble.
We need to see how this is efficiently done in dplyr.