csv averages on linux command line

Question

I have a big file full of records like this

1, 2, 4, 5, 6
1, 3, 5, 6, 3
1, 4, 5, 6, 6
2, 4, 5, 5, 5
2, 3, 4, 5, 2

anyway I need to take the average of all rows with the same 1st number(key). i.e.

1, 3, 4.66, 5.66, 5
2, 3.5, 4.5, 5, 3.5

I know this is something awk/sed would be great for, I just don't have enough experience with them to accomplish it, thanks!

Also, what about averaging those columns together? so, after I output this to a file, id like to get another like:

1, 4.58
1, 4.125

The number of columns to add might not always be 4 either.

EDIT: this might be easier to do in gnuplot, so I mainly just need an answer to the first part.

Answer 1

For the first option:

awk -F, 'BEGIN { OFS=","} {if (!keys[$1]) {keys[$1] = 1}; for (i=2;i<=NF;i++){array[$1,i]+=$i}; count[$1]+=1}END{for (i in keys) {printf ("%s ", i); for (j=2;j<=NF;j++) {printf ("%.2f ", array[i,j]/count[i])}; printf ("%s","\n")}}' inputfile

For the second option:

awk -F, 'BEGIN { OFS=","} {if (!keys[$1]) {keys[$1] = 1}; for (i=2;i<=NF;i++){array[$1,i]+=$i}; count[$1]+=1}END{for (i in keys) {{printf ("%s ", i); sum = 0; for (j=2;j<=NF;j++) {sum += array[i,j]/count[i]}}; printf ("%.2f\n",sum/(NF-1))}}' inputfile

but I'm not sure I understand why you'd want the average of some averages.

Answer 2

This is surprisingly tricky and complicated using Sed, so here's a Python hack to do this:

#!/usr/bin/env python

f = open("mycsv","r")
values = {}
index = {}
for line in f:
    rownum = line.strip().split(", ")
    try:
        values[rownum[0]] = map(lambda x,y: x+y, values[rownum[0]], [float(x) for x in rownum[1:]])
        index[rownum[0]] += 1
    except KeyError:
        values[rownum[0]] = [ float(x) for x in rownum[1:] ]
        index[rownum[0]] = 1

for k,v in values.items():
    values[k] = [x/index[k] for x in values[k]]
    print k, ":", values[k]

This works irrespective of the order of the rows, as long as rows with the same first element are of the same length.

Averaging the columns together will take just one more line of Python in the for loop:

print reduce(lambda x,y: x+y, values[k])/len(values[k])

Given the harrowing number of list comprehensions, though, you're probably better off tackling this problem with NumPy or Matlab.

Answer 3

Karthik has a good suggestion to do this in Numpy: it's just a few lines,

import numpy
data = numpy.loadtxt('filename.txt')
for key in numpy.unique(data.T[0]):
    print data[data.T[0]==key].mean(0)

Or if you wanted to average the columns together, the last line would change to

    avgs = data[data.T[0]==key].mean(0)[1:]
    print avgs[0], avgs[1:].mean()