CODEWITHSUNDEEP
rdplyrtidyr
pachadotdev
pachadotdev

Reputation: 3775

tidyr reshape tibble without expand.grid

I'd like to reshape a tibble without using expand.grid. While expand.grid + delete missing obs + delete "flipped duplicates" (i.e a,b is the same as b,a) shall work, it would be quite slow to compute provided the many combinations I have.

This is a dummy version of what I want to achieve:

library(dplyr)
library(tidyr)

initial_data <- tibble(x = c("east","east","east"), y = c("a","b","c"), z = c(0.1,0.2,0.3))

> initial_data
# A tibble: 3 x 3
      x     y     z
  <chr> <chr> <dbl>
1  east     a   0.1
2  east     b   0.2
3  east     c   0.3

final_data <- tibble(x = c("east","east","east"), y1 = c("a","a","b"), y2 = c("b","c","c"), z1 = c(0.1,0.1,0.2), z2 = c(0.2,0.3,0.3))

> final_data
# A tibble: 3 x 5
      x    y1    y2    z1    z2
  <chr> <chr> <chr> <dbl> <dbl>
1  east     a     b   0.1   0.2
2  east     a     c   0.1   0.3
3  east     b     c   0.2   0.3

This works but is extremely inefficient:

expand_data <- as_tibble(expand.grid(initial_data$x, initial_data$y, initial_data$y)) %>% 
  filter(Var2 != Var3) %>% 
  distinct()

index <- !duplicated(t(apply(expand_data, 1, sort)))
expand_data <- expand_data[index, ] %>% 
  left_join(initial_data, by = c("Var1" = "x", "Var2" = "y")) %>% 
  left_join(initial_data, by = c("Var1" = "x", "Var3" = "y"))

> expand_data
# A tibble: 3 x 5
   Var1  Var2  Var3   z.x   z.y
  <chr> <chr> <chr> <dbl> <dbl>
1  east     b     a   0.2   0.1
2  east     c     a   0.3   0.1
3  east     c     b   0.3   0.2

Many thanks in advance !!

Upvotes: 2

Views: 224

Answers (3)

CPak
CPak

Reputation: 13591

I would give combn a try, combined with purrr::map

Your data

initial_data <- tibble(x = c("east","east","east"), y = c("a","b","c"), z = c(0.1,0.2,0.3))

Solution

initial_data %>%
  nest(-x) %>%
  mutate(data = map(data, ~cbind(as_tibble(t(combn(.x$y, 2))) %>% setNames(paste0("y", 1:2)), 
                        as_tibble(t(combn(initial_data$z, 2))) %>% setNames(paste0("z", 1:2))) )) %>%
  unnest(data)

Output

# A tibble: 3 x 5
      # x    y1    y2    z1    z2
  # <chr> <chr> <chr> <dbl> <dbl>
# 1  east     a     b   0.1   0.2
# 2  east     a     c   0.1   0.3
# 3  east     b     c   0.2   0.3

Upvotes: 1

pogibas
pogibas

Reputation: 28379

Does this base R solution work for you?:

data.frame(x = rep("east", 3),
           matrix(rep(initial_data$y, each = 2), 3), 
           matrix(rep(initial_data$z, each = 2), 3))

#      x X1 X2 X1.1 X2.1
# 1 east  a  b  0.1  0.2
# 2 east  a  c  0.1  0.3
# 3 east  b  c  0.2  0.3

Upvotes: 1

mt1022
mt1022

Reputation: 17299

What about do a inner join and then filter for unique combinations?

library(dplyr)
inner_join(initial_data, initial_data,
           suffix = c('1', '2'), by = 'x') %>%
    filter(y1 < y2) %>%
    select(x, y1, y2, z1, z2)

#      x y1 y2  z1  z2
# 1 east  a  b 0.1 0.2
# 2 east  a  c 0.1 0.3
# 3 east  b  c 0.2 0.3

Upvotes: 1

Related Questions