Greg Rov Greg Rov - 4 months ago 14
Linux Question

Pearson Correlation between two columns

Good Morning. Here is my problem:
I have several files like the one below:

104 0.1697 12.3513214 15.9136214
112 -0.3146 12.0517303 14.8027303
122 0.2718 10.881109 13.259109
123 -0.4185 11.2880142 14.0237142
128 0.0205 13.0585763 15.4365763
132 0.1562 13.3956582 16.9579582
136 -0.4602 12.2567041 14.6347041
157 0.8142 13.6455927 17.2078927
158 -0.9244 8.0012967 11.5635967


Approximately 10000 files, each file with several rows.
And I need to make the Pearson correlation between the column 2 and 4 for each file. Later, I need to make the average of these correlations. And I would like to do everything by Linux commands. Can anyone help me, please?
Thanks

Answer

Try this script. You will need bash and bc (to operate on floating point numbers).

  • give access to execute it chmod +x /path/to/pearson.sh
  • change FILES to your directory where all files are stored
  • call script with no parameters bash /path/to/pearson.sh.

It should produce the mean of all Pearson correlation coefficients calculated on data from those files.

#! /bin/bash

FILES=/path/to/files/

function add {
  echo $1 + $2 | bc
}
function sub {
  echo $1 - $2 | bc
}
function mult {
  echo $1*$2 | bc
}
function div {
  echo $1 / $2 | bc -l
}
function sqrt {
  echo "sqrt ($1)" | bc -l
}

X=0
X2=0
Y=0
Y2=0
XY=0

r=0
R=0
N=0

for f in $FILES/*; do
  N=$((N+1))
  n=0
  while read l; do
    n=$((n+1))
    read -r -a rows <<< $l
    x=${rows[1]}
    y=${rows[3]}
    X=$(add $X $x)
    X2=$(add $X2 $(mult $x $x))
    Y=$(add $Y $y)
    Y2=$(add $Y2 $(mult $y $y))
    XY=$(add $XY $(mult $x $y))
  done < $f;
  r=$(add $r $XY)
  r=$(sub $r $(div $(mult $X $Y) $n))
  d1=$(sub $X2 $(div $(mult $X $X) $n))
  d2=$(sub $Y2 $(div $(mult $Y $Y) $n))
  r=$(div $r $(sqrt $(mult $d1 $d2)))
  R=$(add $R $r)
  X=0
  X2=0
  Y=0
  Y2=0
  XY=0
  r=0
  n=0
done

echo Mean=$(div $R $N)

Ps: I assumed that all files have format like that one you presented. Formula to evaluate the coefficients was taken from the link you gave (http://www.stat.wmich.edu/s216/book/node122.html).

Comments