如何在R编程中将成对相似性写入文件

Question

I have a tab file: string.tab , which contain some think like below

Entry   String
Dog  cube;funny;smart
Cat  tiny;cube;black
....

I also have an package Sim() to compute the similarity between two vectors of string. For example

# here is demo to show how Sim() works
a = c("cs", "funny")
b = c("math", "cool")
score = Sim(a,b)
# output: 0.156

The details of Sim() is not important. It's just a text mining tool.

Here is my code:

data <- read.table("string.tab", sep="\t", header=TRUE)

Now the string.tab in stored in data.

My goal is to compute all pairwise similarity of Entry in string.tab

The output file result should be something like:

Entry1 Entry2 Score
dog cat 0.132
...
...

What is the fast way to do that?

Answer 1

You can use combn(seq(nrow(df)), 2) to get all length 2 combinations (pairs) of row numbers. Then you can apply through these pairs a function which creates a data.frame for that pair, and then rbind the results together.

You can then save this as an R data file with saveRDS , as a CSV with write.csv , etc.

df[] <- lapply(df, as.character)

splits <- lapply(df$String, strsplit, ';')

pairs <- 
apply(utils::combn(seq(nrow(df)), 2), 2, function(x){
  data.frame(Entry1 = df$Entry[x[1]], Entry2 = df$Entry[x[2]], 
             Score = do.call(Sim, splits[x]))
})

pairs.df <- do.call(rbind, pairs)
pairs.df
#   Entry1 Entry2     Score
# 1    Dog    Cat 0.5791888
# 2    Dog  human 0.7434178
# 3    Cat  human 0.4850377

saveRDS(pairs.df, '/path/to/save/file.RDS')
#or
write.csv(pairs.df, '/path/to/save/file.csv')

Data / Functions Used

df <- structure(list(Entry = structure(c(2L, 1L, 3L), .Label = c("Cat", "Dog", "human"), class = "factor"), String = structure(c(1L, 3L, 2L), .Label = c("cube;funny;smart", "man;women", "tiny;cube;black" ), class = "factor")), row.names = c(NA, 3L), class = "data.frame")


Sim <- function(x, y){ # example since I don't have real Sim
  set.seed(sum(nchar(x) + nchar(y)))
  runif(1)
}

Answer 2

The most straight forward thing would be to write a for loop.

I'm going to assume you have some method/function to make a vector from data$String given an index. In this example i'll name the function extract()

 l <- nrow(data)
 n <- choose(l, 2) # number of combinations made
 entry <- data$entry
 result <- data.frame(Entry1 = rep("",n), Entry2 = rep("",n), Score = rep(0,n))
#make combination data
.comb <- data.frame(Entry1 = rep(0,n), Entry2 = rep(0,n))

#Entry1 list
.comb$Entry1 <- unlist(mapply(FUN = rep, x = 1:l, times = (l-1):0))
#Entry2 list
.c <- c(2:l)
if(l>2){
  for(i in 3:l) {
  .c <- c(.c,i:l)
  }
}
.comb$Entry2 <- .c


for(i in 1:n) {
   result[i,"Entry1"] <- data$Entry1[.comb[i,"Entry1"]]
   result[i,"Entry2"] <- data$Entry2[.comb[i,"Entry2"]]
   e.1 <- extract(data$String[.comb[i,"Entry1"]])
   e.2 <- extract(data$String[.comb[i,"Entry2"]])
   result[i, "Score"] <- Sim(e.1,e.2)
}