FR: group_indices() should also return vector of group representatives

[krlmlr]

perhaps as an attribute. This will allow simplifying and accelerating tidyr::nest().

Example:

iris[c(1,2,50,51,100,101), ] %>% group_by(Species) %>% group_indices
## [1] 1 1 1 2 2 3
## attr(,"repr")
## [1] 1 4 6

[romainfrancois]

So the "representative" are 1-based indices of the first row in the data that belong to this group ? @krlmlr.

In that case, we can easily update grouped_indices_grouped_df_impl :

// [[Rcpp::export]]
IntegerVector grouped_indices_grouped_df_impl(GroupedDataFrame gdf) {
  int n = gdf.nrows();
  IntegerVector res = no_init(n);
  int ngroups = gdf.ngroups();
  GroupedDataFrameIndexIterator it = gdf.group_begin();
  for (int i = 0; i < ngroups; i++, ++it) {
    const SlicingIndex& index = *it;
    int n_index = index.size();
    for (int j = 0; j < n_index; j++) {
      res[ index[j] ] = i + 1;
    }
  }
  return res;
}

[romainfrancois]

Need to be careful when doing #341 because empty groups won't have representatives.

They can be NA or 0 or perhaps some unique negative number

[krlmlr]

Good catch. For 1-based representatives, NA_integer_ seems better because df[NA_integer_, ] still is a one-row data frame as opposed to df[0, ]:

iris[NA_integer_, ]
#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> NA           NA          NA           NA          NA    <NA>
iris[0, ]
#> [1] Sepal.Length Sepal.Width  Petal.Length Petal.Width  Species     
#> <0 rows> (or 0-length row.names)

Created on 2018-04-23 by the reprex package (v0.2.0).

[romainfrancois]

We don't need this anymore as we have group_data():

> iris[c(1,2,50,51,100,101), ] %>% group_by(Species) %>% group_data()
# A tibble: 3 x 2
  Species    .rows    
  <fct>      <list>   
1 setosa     <int [3]>
2 versicolor <int [2]>
3 virginica  <int [1]>

tidyr::nest will just use group_data() as in #3622

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/