CODEWITHSUNDEEP
rmatrixcorrelation
CodeNoob
CodeNoob

Reputation: 1840

R: Remove items from a correlation matrix based on matching row and column name

Lets take a look at an example matrix and caclulate the correlations:

some.data <- data.frame(
  A1.1 = c(1,3,4,5,6),
  A1.2 = c(4,5,6,2,3),
  A1.3 = c(3,3,4,2,1),
  A2.1 = c(3,4,5,2,4),
  A2.2 = c(4,5,5,4,2),
  A2.3 = c(1,1,2,2,3),
  A3.1 = c(1,3,4,5,6),
  A3.2 = c(1,4,3,3,4),
  A3.3 = c(4,4,4,4,5)
)
cor.mat <- cor(some.data)

Which gives: 

            A1.1       A1.2       A1.3       A2.1       A2.2       A2.3        A3.1       A3.2       A3.3
A1.1  1.00000000 -0.4109975 -0.6155470 0.06839411 -0.5305954  0.9009862  1.00000000  0.7428336  0.6393620
A1.2 -0.41099747  1.0000000  0.8320503 0.83205029  0.6454972 -0.3779645 -0.41099747  0.0000000 -0.3535534
A1.3 -0.61554702  0.8320503  1.0000000 0.42307692  0.8951436 -0.6289709 -0.61554702 -0.3580574 -0.7844645
A2.1  0.06839411  0.8320503  0.4230769 1.00000000  0.1790287  0.1572427  0.06839411  0.3580574  0.1961161
A2.2 -0.53059545  0.6454972  0.8951436 0.17902872  1.0000000 -0.7319251 -0.53059545 -0.1666667 -0.9128709
A2.3  0.90098616 -0.3779645 -0.6289709 0.15724273 -0.7319251  1.0000000  0.90098616  0.4879500  0.8017837
A3.1  1.00000000 -0.4109975 -0.6155470 0.06839411 -0.5305954  0.9009862  1.00000000  0.7428336  0.6393620
A3.2  0.74283363  0.0000000 -0.3580574 0.35805744 -0.1666667  0.4879500  0.74283363  1.0000000  0.4564355
A3.3  0.63936201 -0.3535534 -0.7844645 0.19611614 -0.9128709  0.8017837  0.63936201  0.4564355  1.0000000

In my original data some columns are dependent, here indicated by the prefixes (A1, A2, A3). As these are not of interest to me I want to set the correlations with the same prefix to zero like so: 

           A1.1       A1.2       A1.3       A2.1       A2.2       A2.3        A3.1       A3.2       A3.3
A1.1          0         0          0   0.06839411 -0.5305954  0.9009862  1.00000000  0.7428336  0.6393620
A1.2          0         0          0   0.83205029  0.6454972 -0.3779645 -0.41099747  0.0000000 -0.3535534
A1.3          0         0          0   0.42307692  0.8951436 -0.6289709 -0.61554702 -0.3580574 -0.7844645
A2.1  0.06839411  0.8320503  0.4230769          0          0          0  0.06839411  0.3580574  0.1961161
A2.2 -0.53059545  0.6454972  0.8951436          0          0          0 -0.53059545 -0.1666667 -0.9128709
A2.3  0.90098616 -0.3779645 -0.6289709          0          0          0  0.90098616  0.4879500  0.8017837
A3.1  1.00000000 -0.4109975 -0.6155470 0.06839411 -0.5305954  0.9009862           0          0          0
A3.2  0.74283363  0.0000000 -0.3580574 0.35805744 -0.1666667  0.4879500           0          0          0 
A3.3  0.63936201 -0.3535534 -0.7844645 0.19611614 -0.9128709  0.8017837           0          0          0

I could do this using a for loop but I suppose this could be done much easier than that?

Upvotes: 0

Views: 857

Answers (2)

akrun
akrun

Reputation: 887148

We could multiply with a block diagonal matrix of 1's

library(Matrix)
as.matrix(cor.mat * !bdiag(replicate(3, matrix(1, 3, 3), simplify = FALSE)))
#        A1.1       A1.2       A1.3       A2.1       A2.2       A2.3        A3.1       A3.2       A3.3
#A1.1  0.00000000  0.0000000  0.0000000 0.06839411 -0.5305954  0.9009862  1.00000000  0.7428336  0.6393620
#A1.2  0.00000000  0.0000000  0.0000000 0.83205029  0.6454972 -0.3779645 -0.41099747  0.0000000 -0.3535534
#A1.3  0.00000000  0.0000000  0.0000000 0.42307692  0.8951436 -0.6289709 -0.61554702 -0.3580574 -0.7844645
#A2.1  0.06839411  0.8320503  0.4230769 0.00000000  0.0000000  0.0000000  0.06839411  0.3580574  0.1961161
#A2.2 -0.53059545  0.6454972  0.8951436 0.00000000  0.0000000  0.0000000 -0.53059545 -0.1666667 -0.9128709
#A2.3  0.90098616 -0.3779645 -0.6289709 0.00000000  0.0000000  0.0000000  0.90098616  0.4879500  0.8017837
#A3.1  1.00000000 -0.4109975 -0.6155470 0.06839411 -0.5305954  0.9009862  0.00000000  0.0000000  0.0000000
#A3.2  0.74283363  0.0000000 -0.3580574 0.35805744 -0.1666667  0.4879500  0.00000000  0.0000000  0.0000000
#A3.3  0.63936201 -0.3535534 -0.7844645 0.19611614 -0.9128709  0.8017837  0.00000000  0.0000000  0.0000000

Or another option is to use row/column index

replace(cor.mat, cbind(rep(1:9, each = 3),
       c(sapply(list(1:3, 4:6, 7:9), rep, 3))), 0)

Or use outer to construct a logical matrix and multiply with the cor.mat

nm1 <- sub("\\.\\d+$", "", colnames(cor.mat))
cor.mat * outer(nm1, nm1, `!=`)

Upvotes: 2

markus
markus

Reputation: 26343

One option is to reshape your data from wide to long such that it will contain three columns

cor.mat_long <- reshape2::melt(cor.mat)
cor.mat_long
#   Var1 Var2       value
#1  A1.1 A1.1  1.00000000
#2  A1.2 A1.1 -0.41099747
#3  A1.3 A1.1 -0.61554702
#4  A2.1 A1.1  0.06839411
#5  A2.2 A1.1 -0.53059545
#6  A2.3 A1.1  0.90098616
#...

Create a logical vector based on the prefixes of Var1 and Var2 that indicates when these prefixes are the same. Use this vector to replace cor.mat_long$value with 0 where it evaluates to TRUE

cor.mat_long$value[with(cor.mat_long, sub("\\.\\d+$", "", Var1) == sub("\\.\\d+$", "", Var2))] <- 0

Finally reshape to wide format again.

cor.mat2 <- reshape2::dcast(cor.mat_long, Var1 ~ Var2)
cor.mat2
#  Var1        A1.1       A1.2       A1.3       A2.1       A2.2       A2.3        A3.1       A3.2       A3.3
#1 A1.1  0.00000000  0.0000000  0.0000000 0.06839411 -0.5305954  0.9009862  1.00000000  0.7428336  0.6393620
#2 A1.2  0.00000000  0.0000000  0.0000000 0.83205029  0.6454972 -0.3779645 -0.41099747  0.0000000 -0.3535534
#3 A1.3  0.00000000  0.0000000  0.0000000 0.42307692  0.8951436 -0.6289709 -0.61554702 -0.3580574 -0.7844645
#4 A2.1  0.06839411  0.8320503  0.4230769 0.00000000  0.0000000  0.0000000  0.06839411  0.3580574  0.1961161
#5 A2.2 -0.53059545  0.6454972  0.8951436 0.00000000  0.0000000  0.0000000 -0.53059545 -0.1666667 -0.9128709
#6 A2.3  0.90098616 -0.3779645 -0.6289709 0.00000000  0.0000000  0.0000000  0.90098616  0.4879500  0.8017837
#7 A3.1  1.00000000 -0.4109975 -0.6155470 0.06839411 -0.5305954  0.9009862  0.00000000  0.0000000  0.0000000
#8 A3.2  0.74283363  0.0000000 -0.3580574 0.35805744 -0.1666667  0.4879500  0.00000000  0.0000000  0.0000000
#9 A3.3  0.63936201 -0.3535534 -0.7844645 0.19611614 -0.9128709  0.8017837  0.00000000  0.0000000  0.0000000

If you don't want Var1 as an explicit column, do

rownames(cor.mat2) <- cor.mat2$Var1
cor.mat2 <- cor.mat2[-1]

Don't know if this is much easier than your loop though.

Upvotes: 3

Related Questions