gapply: grouped ordered apply

In WinVector/replyr: Patches to Use 'dplyr' on Remote Data Sources

Description

Partitions from by values in grouping column, applies a generic transform to each group and then binds the groups back together. Only advised for a moderate number of groups and better if grouping column is an index. This is powerful enough to implement "The Split-Apply-Combine Strategy for Data Analysis" https://www.jstatsoft.org/article/view/v040i01

Usage

gapply(
  df,
  gcolumn,
  f,
  ...,
  ocolumn = NULL,
  decreasing = FALSE,
  partitionMethod = "split",
  bindrows = TRUE,
  maxgroups = 100,
  eagerCompute = FALSE,
  restoreGroup = FALSE,
  tempNameGenerator = mk_tmp_name_source("replyr_gapply")
)

Arguments

df	remote dplyr data item
gcolumn	grouping column
f	transform function or pipeline
...	force later values to be bound by name
ocolumn	ordering column (optional)
decreasing	logical, if TRUE sort in decreasing order by ocolumn
partitionMethod	method to partition the data, one of 'group_by' (depends on f being dplyr compatible), 'split' (only works over local data frames), or 'extract'
bindrows	logical, if TRUE bind the rows back into a data item, else return split list
maxgroups	maximum number of groups to work over (intentionally not enforced if partitionMethod=='group_by')
eagerCompute	logical, if TRUE call compute on split results
restoreGroup	logical, if TRUE restore group column after apply when partitionMethod %in% c('extract', 'split')
tempNameGenerator	temp name generator produced by wrapr::mk_tmp_name_source, used to record dplyr::compute() effects.

Details

Note this is a fairly expensive operator, so it only makes sense to use in situations where f itself is fairly complicated and/or expensive.

Value

transformed frame

Examples

d <- data.frame(
  group = c(1, 1, 2, 2, 2),
  order = c(.1, .2, .3, .4, .5),
  values = c(10, 20, 2, 4, 8)
)

# User supplied window functions.  They depend on known column names and
# the data back-end matching function names (as cumsum).
cumulative_sum <- function(d) {
  dplyr::mutate(d, cv = cumsum(values))
}
rank_in_group <- function(d) {
  d %.>%
    dplyr::mutate(., constcol = 1) %.>%
    dplyr::mutate(., rank = cumsum(constcol)) %.>%
    dplyr::select(., -constcol)
}

for (partitionMethod in c('group_by', 'split', 'extract')) {
  print(partitionMethod)
  print('cumulative sum example')
  print(
    gapply(
      d,
      'group',
      cumulative_sum,
      ocolumn = 'order',
      partitionMethod = partitionMethod
    )
  )
  print('ranking example')
  print(
    gapply(
      d,
      'group',
      rank_in_group,
      ocolumn = 'order',
      partitionMethod = partitionMethod
    )
  )
  print('ranking example (decreasing)')
  print(
    gapply(
      d,
      'group',
      rank_in_group,
      ocolumn = 'order',
      decreasing = TRUE,
      partitionMethod = partitionMethod
    )
  )
}

WinVector/replyr documentation built on Oct. 22, 2020, 8:07 p.m.