Using a self-written function inside group_by() and mutate()

tidyverse
1

Hi all

I am trying to get a function I have written for a single dataset to work inside a group_by() mutate() call, but I currently get an error with regards to the number of rows in the group.

I have made a fully reproducible reprex (first one ever, is awesome) regarding my issue. The function works fine on a single dataset and within a group_by() and a do() but not within group_by() mutate(). Any tips or ideas why are much appreciated!

# I am trying to calculate the cumulative distance between latitude and longitude points within a group_by call. For example I have latitude and longitude points for multiple runs I have done and would like to get the cumulative distance for each one.

# load packages
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)

# custom function for getting distances between latitude and longitude points
# REQUIRES GEOSPHERE
get_dists <- function(dat_in, lon = 'lon', lat = 'lat'){
  
  dat <- dat_in[,c(lon, lat)]
  names(dat) <- c('lon', 'lat')
  
  out <- sapply(2:nrow(dat), function(y){geosphere::distm(dat[y-1,], dat[y,])/1000})
  out <-    c(0, cumsum(out))
  return(out)
}

# create fake data
d <- data.frame(run = c(1, 1, 1, 2, 2, 2),
                lat = c(57.15508, 57.15521, 57.15520, 52.41278, 52.41283, 52.41317),
                lon = c(-2.07886, -2.07886, -2.07887, -4.07803, -4.07806, -4.07858))

# calculate distance between all points irrespective of run
get_dists(d)
#> [1]   0.00000000   0.01447153   0.01573792 543.25211631 543.25804333
#> [6] 543.30980443

# calculate distance between points grouped_by run
d %>% 
  group_by(run) %>%
  mutate(dists = get_dists(.))
#> Error in mutate_impl(.data, dots): Column `dists` must be length 3 (the group size) or one, not 6

# i get an error I have not managed to fix. Think it is because I have a data argument in my distance function but I do not know how to solve it.

# however it does work with a do function
d %>% 
  group_by(run) %>%
  do(data.frame(lat = .$lat,
                lon = .$lon,
                dists = get_dists(.)))
#> # A tibble: 6 x 4
#> # Groups:   run [2]
#>     run      lat      lon       dists
#>   <dbl>    <dbl>    <dbl>       <dbl>
#> 1     1 57.15508 -2.07886 0.000000000
#> 2     1 57.15521 -2.07886 0.014471534
#> 3     1 57.15520 -2.07887 0.015737917
#> 4     2 52.41278 -4.07803 0.000000000
#> 5     2 52.41283 -4.07806 0.005927023
#> 6     2 52.41317 -4.07858 0.057688123

# any idea what is going on?
2 Likes
2

In this part

d %>% 
  group_by(run) %>%
  mutate(dists = get_dists(.))

I think the . is understood as the %>% one so it is replaced by the all data.frame d not just the subset group data. It is why the dimension is 6 not 3. mutate works well with vector.

With your current function taking a data.frame, you can use tidyr::nest to create a list column with grouped data and apply you function with map on the list -column

library(dplyr, warn.conflicts = F)
library(tidyr)

# custom function for getting distances between latitude and longitude points
# REQUIRES GEOSPHERE
get_dists <- function(dat_in, lon = 'lon', lat = 'lat'){
  
  dat <- dat_in[,c(lon, lat)]
  names(dat) <- c('lon', 'lat')
  
  out <- sapply(2:nrow(dat), function(y){geosphere::distm(dat[y-1,], dat[y,])/1000})
  out <-    c(0, cumsum(out))
  return(out)
}

# create fake data
d <- data_frame(run = c(1, 1, 1, 2, 2, 2),
                lat = c(57.15508, 57.15521, 57.15520, 52.41278, 52.41283, 52.41317),
                lon = c(-2.07886, -2.07886, -2.07887, -4.07803, -4.07806, -4.07858))

d %>% 
  nest(-run) %>%
  group_by(run) %>%
  mutate(dists = purrr::map(data, get_dists)) %>%
  unnest()
#> # A tibble: 6 x 4
#> # Groups:   run [2]
#>     run       dists      lat      lon
#>   <dbl>       <dbl>    <dbl>    <dbl>
#> 1     1 0.000000000 57.15508 -2.07886
#> 2     1 0.014471534 57.15521 -2.07886
#> 3     1 0.015737917 57.15520 -2.07887
#> 4     2 0.000000000 52.41278 -4.07803
#> 5     2 0.005927023 52.41283 -4.07806
#> 6     2 0.057688123 52.41317 -4.07858

However, you can also skip the nesting part by recreating the data.frame your function work with

d %>% 
  group_by(run) %>%
  mutate(dists = tibble(lat, lon) %>% get_dists)
#> # A tibble: 6 x 4
#> # Groups:   run [2]
#>     run      lat      lon       dists
#>   <dbl>    <dbl>    <dbl>       <dbl>
#> 1     1 57.15508 -2.07886 0.000000000
#> 2     1 57.15521 -2.07886 0.014471534
#> 3     1 57.15520 -2.07887 0.015737917
#> 4     2 52.41278 -4.07803 0.000000000
#> 5     2 52.41283 -4.07806 0.005927023
#> 6     2 52.41317 -4.07858 0.057688123

or change your function to take two vectors so that it's work better with grouped-by mutate

get_dists2 <- function(lon, lat){
  
  dat <- tibble(lon, lat)
  names(dat) <- c('lon', 'lat')
  
  out <- sapply(2:nrow(dat), function(y){geosphere::distm(dat[y-1,], dat[y,])/1000})
  out <-    c(0, cumsum(out))
  return(out)
}

d %>% 
  group_by(run) %>%
  mutate(dists = get_dists2(lon, lat))
#> # A tibble: 6 x 4
#> # Groups:   run [2]
#>     run      lat      lon       dists
#>   <dbl>    <dbl>    <dbl>       <dbl>
#> 1     1 57.15508 -2.07886 0.000000000
#> 2     1 57.15521 -2.07886 0.014471534
#> 3     1 57.15520 -2.07887 0.015737917
#> 4     2 52.41278 -4.07803 0.000000000
#> 5     2 52.41283 -4.07806 0.005927023
#> 6     2 52.41317 -4.07858 0.057688123
7 Likes
3

All of these solutions are amazing. Thank you so much.