How to “flatten” or “collapse” a 2D data frame into a 1D data frame in R?

Question

I have a two dimensional table with distances in a data.frame in R (imported from csv):

           CP000036   CP001063      CP001368
CP000036      0           a            b
CP001063      a           0            c
CP001368      b           c            0

I'd like to "flatten" it. that I have one axes's value in the first col, and the other axes's value in the second col, and then the distance in the third col:

Genome1      Genome2       Dist
CP000036     CP001063       a
CP000036     CP001368       b
CP001063     CP001368       c

Above is ideal, but it would be completely fine to have repetition such that each cell in the input matrix has it's own row:

Genome1      Genome2       Dist
CP000036     CP000036       0
CP000036     CP001063       a
CP000036     CP001368       b
CP001063     CP000036       a
CP001063     CP001063       0
CP001063     CP001368       c
CP001368     CP000036       b
CP001368     CP001063       c
CP001368     CP001368       0

Here is an example 3x3 matrix, but my dataset I is much larger (about 2000x2000). I would do this in Excel, but I need ~3 million rows for the output, whereas Excel's maximum is ~1 million.

This question is very similar to "How to “flatten” or “collapse” a 2D Excel table into 1D?"1

Answer 1

So this is one solution using melt from the package reshape2:

dm <- 
  data.frame( CP000036 = c( "0", "a", "b" ),
              CP001063 = c( "a", "0", "c" ),
              CP001368 = c( "b", "c", "0" ),
              stringsAsFactors = FALSE,
              row.names = c( "CP000036", "CP001063", "CP001368" ) )

# assuming the distance follows a metric we avoid everything below and on the diagonal
dm[ lower.tri( dm, diag = TRUE ) ]  <- NA
dm$Genome1 <- rownames( dm )

# finally melt and avoid the entries below the diagonal with na.rm = TRUE
library(reshape2) 
dm.molten <- melt( dm, na.rm= TRUE, id.vars="Genome1",
                   value.name="Dist", variable.name="Genome2" )

print( dm.molten )
   Genome1  Genome2 Dist
4 CP000036 CP001063    a
7 CP000036 CP001368    b
8 CP001063 CP001368    c

Probably there are more performant solutions but I like this one because it's plain and simple.