dplyr: R crash/errors with group_by, mutate - r

Question

I'm writing a function to remove duplicate observations of undirected relationships between firms when both parties report the relationship. For my ~1.3 million observation dataset, the function collapse_undirected below results either in R crashing during the mutate for hash or the error:
"translateCharUTF8 must be called on CHARSXP"
during the mutate for dup.
The goal of this function is to uniquely identify each pair of related firms by ordering and concatenating their IDs, and then dropping duplicate hashes reported in the same time period.
The data set I'm using is licensed so I can't provide it, but the "translateCharUTF8" error is reproducible with the randomly generated data I've included below. It occurs more frequently with larger sets. I'd say its common at around 9000 observations. I've also included a slow version of the function that runs without incident, which further leads me to believe that the problem occurs in the first mutate.
The function with the error:
collapse_undirected <- function(data, dir){
out <- data %>% filter(REL_TYPE != dir)
obs <- data %>% filter(REL_TYPE == dir) %>%
group_by(SOURCE, TARGET) %>%
mutate(hash = paste(min(SOURCE, TARGET),max(SOURCE, TARGET))) %>%
group_by(START, END) %>%
mutate(dup = duplicated(hash)) %>%
filter(!dup) %>%
select(-hash,-dup)
bind_rows(out,obs)
}
The slow workaround:
jank_undir <- function(data, dir){
obs <- data %>% filter(REL_TYPE == dir)
out <- data %>% filter(REL_TYPE != dir)
obs$hash <- NA
for(i in 1:nrow(obs)){
obs$hash[i] <- paste(min(obs$SOURCE[i], obs$TARGET[i]),
max(obs$SOURCE[i], obs$TARGET[i]))
}
obs %>% group_by(START,END) %>%
mutate(dup = duplicated(hash)) %>%
filter(!dup) %>%
select(-hash,-dup) %>%
bind_rows(out)
}
Here's a convenience function to randomly generate test data:
reroll <- function(n){
test_data <- data_frame(1:n)
test_data$SOURCE <- as.character(sample(1:27000, size = n, replace = TRUE))
test_data$TARGET <- as.character(sample(1:27000, size = n, replace = TRUE))
test_data$REL_TYPE <- "DUMMY"
test_data$START <- sample(1:2870, size = n, replace = TRUE)
test_data$END <- sample(1:2781, size = n, replace = TRUE)
test_data
}
And, varying with the random draw, this should demonstrate the error:
library(dplyr)
test_data <- reroll(9000)
test_cleaned <- test_data %>% jank_undir("DUMMY")
test_cleaned <- test_data %>% collapse_undirected("DUMMY")
I'd greatly appreciate any insight into why this is happening. The slow version is fast enough for now but I anticipate needing to use it for larger datasets. The R crashes occurred on both my Windows and Linux based R sessions with the main dataset but seems less frequent on the Linux version. My dplyr is 0.7.2
Thank you,