CODEWITHSUNDEEP
rloopsmatrix
Edward
Edward

Reputation: 4623

Double loop to fill a correlation matrix

I have a dataset like this

set.seed(1)
a = abs(rnorm(10, mean = 0, sd= 1))
b = abs(rnorm(10, mean = 0, sd= 1))
c = abs(rnorm(10, mean = 0, sd= 1))
d = abs(rnorm(10, mean = 0, sd= 1))
df = as.data.frame(cbind(a, b, c, d))

And i want to get a table 

   c   d
a 0.5 0.1
b 0.8 0.3

where cols and rows are the variables and cells - correlation coefficients between variables.

I do as below

for(j in df[, 1:2])           {
for(i in df[, 3:4]) {

  k=abs(cor.test(j, i, method = c( "spearman"))$estimate)
  cat(k, '\n')
  y <- rbind(y, k)
}}
y

and get 

rho
k 0.175757576
k 0.006060606
k 0.151515152
k 0.054545455

I used this post Using double loop to fill a matrix in R

mat<-matrix(list(c(NA,NA)), nrow=2, ncol=2)
for(j in df[, 1:2])           {
  for(i in df[, 3:4]) {

    mat[i,j][[1]]=abs(cor.test(j, i, method = c( "spearman"))$estimate)

  }}
mat

and i get

     [,1]      [,2]     
[1,] Logical,2 Logical,2
[2,] Logical,2 Logical,2

how to fill the table? Or can i fill it without loop?

Upvotes: 3

Views: 1075

Answers (2)

jogo
jogo

Reputation: 12559

The function cor()can do that:

set.seed(1)
a = abs(rnorm(10, mean = 0, sd= 1))
b = abs(rnorm(10, mean = 0, sd= 1))
c = abs(rnorm(10, mean = 0, sd= 1))
d = abs(rnorm(10, mean = 0, sd= 1))
#### df = as.data.frame(cbind(a, b, c, d)) # not used
cor(cbind(a,b), cbind(c,d))
# > cor(cbind(a,b), cbind(c,d))
#           c          d
# a 0.5516642 -0.3918783
# b 0.8200195  0.1474773

And you can do abs() for your desired result:

abs(cor(cbind(a,b), cbind(c,d)))
# > abs(cor(cbind(a,b), cbind(c,d)))
# c         d
# a 0.5516642 0.3918783
# b 0.8200195 0.1474773

Spearman:

abs(cor(cbind(a,b), cbind(c,d), method = "spearman"))
# > abs(cor(cbind(a,b), cbind(c,d), method = "spearman"))
# c          d
# a 0.05454545 0.40606061
# b 0.75757576 0.05454545

If you want to use your dataframe, you can do:

df = as.data.frame(cbind(a, b, c, d))
rm(a,b,c,d) ### to be sure that a, ..., d are from the dataframe.
with(df, abs(cor(cbind(a,b), cbind(c,d), method = "spearman")))

or

abs(cor(df[,c("a", "b")], df[,c("c","d")], method = "spearman"))

Upvotes: 1

d.b
d.b

Reputation: 32538

I would compute correlation matrix for df one time and then subset whatever combinations I need from that. This way, you wouldn't have to run cor multiple times.

m = cor(df, method = "spearman")
m[row.names(m) %in% c("a","b"), colnames(m) %in% c("c","d")]
#           c           d
#a 0.05454545 -0.40606061
#b 0.75757576  0.05454545

Upvotes: 2

Related Questions