efficient way finding pair value in adjacent columns (Python/R/Sql)

Question

I have a dataframe( df below) in pandas with several million rows * 20 columns.

And given a pair of values, I'm trying to find if they exist in adjacent columns of df .

eg df looks like

Given pairs of value (a3, b2) , we find they exist in adjacent columns(doesn't have to be in the same row).

For pairs of value (b2, a3) , we don't consider they meet the condition(the shift of column should be to the right).

This can be done using loops for a small dataset, but I have millions of records with 20 columns and lots of pairs to check. Is there any way to compute this efficiently? Thanks!

Answer 1

Here is a base R solution by defining a function isAdjacent

isAdjacent <- function(df,p) {
    colnum <- col(df)
    diff(sapply(p,function(x) colnum[df==x],USE.NAMES = FALSE))==1
}

where df is the data.frame, and p is the pair.

Example

p1 <- c("b1","c2")
p2 <- rev(p1)
p3 <- c("a1","c3")

> isAdjacent(df,p1)
[1] TRUE

> isAdjacent(df,p2)
[1] FALSE

> isAdjacent(df,p3)
[1] FALSE

Data

> dput(df)
structure(list(A = c("a1", "a2", "a3", "a4"), B = c("b1", "b2", 
"b3", "b4"), C = c("c1", "c2", "c3", "c4"), D = c("d1", "d2", 
"d3", "d4"), E = c("e1", "e2", "e3", "e4"), F = c("f1", "f2",
"f3", "f4"), G = c("g1", "g2", "g3", "g4")), class = "data.frame", row.names = c(NA, 

-4L))

Large Data Example (Benchmarking)

df <- setNames(as.data.frame(sapply(letters[1:20], paste0, 1:1e6)), LETTERS[1:20])

p <- c("a1", "c3")
system.time({
    isAdjacent <- function(df, p) {
        colnum <- col(df)
        diff(sapply(p, function(x) colnum[df == x], USE.NAMES = FALSE)) == 1
    }
    isAdjacent(df, p)
})
#   user  system elapsed 
#   1.03    0.07    1.11

library(data.table)
system.time({
    DT <- data.table(VAL = unlist(df), COL = rep(1L:ncol(df), each = nrow(df)), key = "VAL")
    isadj <- function(left, right) {
        DT[.(left), .(COL = COL + 1L)][DT[.(right)], on = .(COL), nomatch = 0L, .N > 0L]
    }
    isadj(p[1], p[2])
})

#   user  system elapsed
#  35.79    1.91   36.24

Answer 2

Using df in ThomasIsCoding's post, here is an option using data.table in R:

library(data.table)
DT <- data.table(VAL=unlist(df), COL=rep(1L:ncol(df), each=nrow(df)), key="VAL")
isadj <- function(left, right) {
    DT[.(left), .(COL=COL+1L)][DT[.(right)], on=.(COL), nomatch=0L, .N > 0L]
}

isadj("a3", "b2")    
#[1] TRUE

isadj("b2", "a3")    
#[1] FALSE