Stepwise sampling in a tibble

user_6412795 0
Viewed 53

I am trying to simulate some data by sampling multiple steps.

The first step (create x) works fine.

In the second step, I want to create the variable y by sampling from different vectors based on the value of x.

My code runs without errors, but fails at what I am trying to achieve as it only samples one value for e.g., x == "A", and then reuses that value for all subsequent rows where x == "A". I want it to sample one time for each row where x == "A"

Code:

library(tidyverse)
set.seed(1)

data <- tibble(
  x = sample(c("A", "B", "C"), size = 10000, prob = c(0.1, 0.2, 0.7), replace = TRUE),
  y = case_when(
    x == "A" ~ sample(c("A1", "A2", "A3"), size = 1, prob = c(0.3, 0.4, 0.3)),
    x == "B" ~ sample(c("B1", "B2", "B3"), size = 1, prob = c(0.3, 0.4, 0.3)),
    x == "C" ~ sample(c("C1", "C2", "C3"), size = 1, prob = c(0.3, 0.4, 0.3)),
  ))

unique(data$x)
[1] "C" "A" "B"

unique(data$y)
[1] "C1" "A2" "B3"

If the code works as intended unique(data$y) should return something similar to [1] "A1", "A2", "A3", "B1", "B2", "B3", "C1", "C2", "C3"

I know the problem is the size = 1 argument in sample(), but what can I replace it with? Removing it returns the error:

Error: `x == "A" ~ sample(c("A1", "A2", "A3"), prob = c(0.3, 0.4, 0.3))` must be length 100 or one, not 3

And I have tried size = nrow(.data) and size=nrow(.), but that also returns error.

Is there a simple solution to this?

tibble dplyr r random
3 Answers

There maybe a simpler way but this stays close to your original code and works...

data <- tibble(
  x = sample(c("A", "B", "C"), size = 10000, prob = c(0.1, 0.2, 0.7), replace = TRUE)) %>%
  rowwise() %>%
  summarise(x= x, 
            y = case_when(
    x == "A" ~ sample(c("A1", "A2", "A3"), size = 1, prob = c(0.3, 0.4, 0.3)),
    x == "B" ~ sample(c("B1", "B2", "B3"), size = 1, prob = c(0.3, 0.4, 0.3)),
    x == "C" ~ sample(c("C1", "C2", "C3"), size = 1, prob = c(0.3, 0.4, 0.3)),
  ))
user_15274030

It has to do with vectorized functions and recycling. If you do it vectorized, it will recycle the same value. If you do it with a loop it works. For example, 

v1 <- c('A', 'A', 'B', 'B', 'C', 'C', 'C', 'A', 'A')

#Vectorized ifelse
ifelse(v1 == 'A', sample(c("A1", "A2", "A3"), size = 1, prob = c(0.3, 0.4, 0.3)), NA)
#[1] "A3" "A3" NA   NA   NA   NA   NA   "A3" "A3"

#Not vectorized if/else with a loop,
sapply(v1, function(i) if (i == 'A') { sample(c("A1", "A2", "A3"), size = 1, prob = c(0.3, 0.4, 0.3)) }else {NA})
#   A    A    B    B    C    C    C    A    A 
#"A2" "A3"   NA   NA   NA   NA   NA "A2" "A1"
user_56355800

It would be simple to understand if you divide this into steps

library(dplyr)
data <- tibble(
   x = sample(c("A", "B", "C"), size = 10000, 
                prob = c(0.1, 0.2, 0.7), replace = TRUE))

data <- data %>%
  mutate(y = case_when(
     x == "A" ~ sample(c("A1", "A2", "A3"), size = n(), 
               prob = c(0.3, 0.4, 0.3), replace = TRUE),
     x == "B" ~ sample(c("B1", "B2", "B3"), size = n(), 
                 prob = c(0.3, 0.4, 0.3), replace = TRUE),
     x == "C" ~ sample(c("C1", "C2", "C3"), size = n(), 
                prob = c(0.3, 0.4, 0.3), replace = TRUE),
)) 

unique(data$y)
#[1] "C2" "B3" "A1" "C3" "B1" "C1" "B2" "A3" "A2"

Or if you want to continue your way you need to specify the size argument same as the one mentioned for x with replace = TRUE

data <- tibble(
  x = sample(c("A", "B", "C"), size = 10000, 
            prob = c(0.1, 0.2, 0.7), replace = TRUE),
  y = case_when(
    x == "A" ~ sample(c("A1", "A2", "A3"), size = 10000, 
                  prob = c(0.3, 0.4, 0.3), replace = TRUE),
    x == "B" ~ sample(c("B1", "B2", "B3"), size = 10000, 
                  prob = c(0.3, 0.4, 0.3), replace = TRUE),
    x == "C" ~ sample(c("C1", "C2", "C3"), size = 10000, 
                  prob = c(0.3, 0.4, 0.3), replace = TRUE),
  ))
user_39629140
Related