Computing minimum distance between a row and all previous rows in R - r

Question

I want to compute the minimum distance between the current row and every row before it within each group. My data frame has several groups, and each group has multiple dates with longitude and latitude. I use a Haversine function to compute distance, and I need to apply this function as described above. The data frame looks like the following:
grp date long lat rowid
1 1 1995-07-01 11 12 1
2 1 1995-07-05 3 0 2
3 1 1995-07-09 13 4 3
4 1 1995-07-13 4 25 4
5 2 1995-03-07 12 6 1
6 2 1995-03-10 3 27 2
7 2 1995-03-13 34 8 3
8 2 1995-03-16 25 9 4
My current attempt uses purrrlyr::by_row, but the method is too slow. In practice, each group has thousands of dates and geographic positions. Here is part of my current attempt:
calc_min_distance <- function(df, grp.name, row){
df %>%
filter(
group_name==grp.name
) %>%
filter(
row_number() <= row
) %>%
mutate(
last.lat = last(lat),
last.long = last(long),
rowid = 1:n()
) %>%
group_by(rowid) %>%
purrrlyr::by_row(
~haversinedistance.fnct(.$last.long, .$last.lat, .$long, .$lat),
.collate='rows',
.to = 'min.distance'
) %>%
filter(
row_number() < n()
) %>%
summarise(
min = min(min.distance)
) %>%
.$min
}
df_dist <-
df %>%
group_by(grp_name) %>%
mutate(rowid = 1:n()) %>%
group_by(grp_name, rowid) %>%
purrrlyr::by_row(
~calc_min_distance(df, .$grp_name,.$rowid),
.collate='rows',
.to = 'min.distance'
) %>%
ungroup %>%
select(-rowid)
Suppose that distance is defined as (lat + long) for reference row - (lat + long) for each pairwise row less than the reference row. My expected output for grp 1 is the following:
grp date long lat rowid min.distance
1 1 1995-07-01 11 12 1 0
2 1 1995-07-05 3 0 2 -20
3 1 1995-07-09 13 4 3 -6
4 1 1995-07-13 4 25 4 6
How can I quickly compute the minimum distance between the current rowid and all rowids before it?

Answer 1

Here's how I would go about it. You need to calculate all the within-group pair-wise distances anyway, so we'll use geosphere::distm which is designed to do just that. I'd suggest stepping through my function line-by-line and looking at what it does, I think it will make sense.
library(geosphere)
find_min_dist_above = function(long, lat, fun = distHaversine) {
d = distm(x = cbind(long, lat), fun = fun)
d[lower.tri(d, diag = TRUE)] = NA
d[1, 1] = 0
return(apply(d, MAR = 2, min, na.rm = TRUE))
}
df %>% group_by(grp) %>%
mutate(min.distance = find_min_dist_above(long, lat))
# # A tibble: 8 x 6
# # Groups: grp [2]
# grp date long lat rowid min.distance
# <int> <fct> <int> <int> <int> <dbl>
# 1 1 1995-07-01 11 12 1 0
# 2 1 1995-07-05 3 0 2 1601842.
# 3 1 1995-07-09 13 4 3 917395.
# 4 1 1995-07-13 4 25 4 1623922.
# 5 2 1995-03-07 12 6 1 0
# 6 2 1995-03-10 3 27 2 2524759.
# 7 2 1995-03-13 34 8 3 2440596.
# 8 2 1995-03-16 25 9 4 997069.
Using this data:
df = read.table(text = ' grp date long lat rowid
1 1 1995-07-01 11 12 1
2 1 1995-07-05 3 0 2
3 1 1995-07-09 13 4 3
4 1 1995-07-13 4 25 4
5 2 1995-03-07 12 6 1
6 2 1995-03-10 3 27 2
7 2 1995-03-13 34 8 3
8 2 1995-03-16 25 9 4', h = TRUE)