AWK: statistics operations of multi-column CSV data - csv

Question

With the aim to perform some statistical analysis of multi-column data I am analyzing big number of CSV filles using the following bash + AWK routine:
#!/bin/bash
home="$PWD"
# folder with the outputs
rescore="${home}"/rescore
# folder with the folders to analyse
storage="${home}"/results
#cd "${home}"/results
cd ${storage}
csv_pattern='*_filt.csv'
while read -r d; do
awk -v rescore="$rescore" '
FNR==1 {
if (n)
mean[suffix] = s/n
prefix=suffix=FILENAME
sub(/_.*/, "", prefix)
sub(/\/[^\/]+$/, "", suffix)
sub(/^.*_/, "", suffix)
s=n=0
}
FNR > 1 {
s += $3
++n
}
END {
out = rescore "/" prefix ".csv"
mean[suffix] = s/n
print prefix ":", "dG(mean)" > out
for (i in mean)
printf "%s: %.2f\n", i, mean[i] >> out
close(out)
}' "${d}_"*/${csv_pattern} #> "${rescore}/"${d%%_*}".csv"
done < <(find . -maxdepth 1 -type d -name '*_*_*' | awk -F '[_/]' '!seen[$2]++ {print $2}')
Basically the script takes ensemble of CSV files belonged to the same prefix (defined as the naming pattern occured at the begining of the directory contained CSV, for example 10V1 from 10V1_cne_lig1) and calculate for it the mean value for the numbers in the third column:
# input *_filt.csv located in the folder 10V1_cne_lig1001
ID, POP, dG
1, 142, -5.6500
2, 10, -5.5000
3, 2, -4.9500
add 1 string to 10V1.csv, which is organized in 2 column format i) the name of the suffix of the folder with initial CSV; ii) the mean value calculated for all numbers in the third column (dG) of input.csv:
# this is two column format of output.csv: 10V1.csv
10V1: dG(mean)
lig1001: -5.37
in this way for 100 CSV filles such output.csv should contain 100 lines with the mean values, etc
I need to introduce a small modification to my AWK part of my routine that would add the 3rd column to the output CSV with RMSD value (as the measure of the differences between initial dG values) of the initial data (dG), which had been used to calculate the MEAN value. Using AWK syntax, with a particular MEAN value the RMS could be expressed as
mean=$(awk -F , '{sum+=$3}END{printf "%.2f", sum/NR}' $csv)
rmsd=$(awk -v mean=$mean '{++n;sum+=($NF-mean)^2} END{if(n) printf "%.2f", sqrt(sum/n)}' $csv)
Here is expected output for 5 means and 5 rmsds values calculated for 5 CSV logs (the first one is corresponded to my above example!):
10V1: dG(mean): RMSD (error)
lig1001 -5.37 0.30
lig1002 -8.53 0.34
lig1003 -6.57 0.25
lig1004 -9.53 0.00 # rmsd=0 since initial csv has only 1 line: no data variance
lig1005 -8.11 0.39
How this addition could be incorporated into my main bash-AWK code with the aim to add the third RMSD column (for each of the processed CSV, thus taking each of the calculated MEAN) to the output.csv?

Answer 1

You can calculate both of mean and rmsd within the awk code.
Would you please try the following awk code:
awk -v rescore="$rescore" '
FNR==1 {
if (n) { # calculate the results of previous file
m = s / n # mean
var = s2 / n - m * m # variance
if (var < 0) var = 0 # avoid an exception due to round-off error
mean[suffix] = m # store the mean in an array
rmsd[suffix] = sqrt(var)
}
prefix=suffix=FILENAME
sub(/_.*/, "", prefix)
sub(/\/[^\/]+$/, "", suffix)
sub(/^.*_/, "", suffix)
s = 0 # sum of $3
s2 = 0 # sum of $3 ** 2
n = 0 # count of samples
}
FNR > 1 {
s += $3
s2 += $3 * $3
++n
}
END {
out = rescore "/" prefix ".csv"
m = s / n
var = s2 / n - m * m
if (var < 0) var = 0
mean[suffix] = m
rmsd[suffix] = sqrt(var)
print prefix ":", "dG(mean)", "dG(rmsd)" > out
for (i in mean)
printf "%s: %.2f %.2f\n", i, mean[i], rmsd[i] >> out
close(out)
}'
Here is the version to print the lowest value of dG.
awk -v rescore="$rescore" '
FNR==1 {
if (n) { # calculate the results of previous file
m = s / n # mean
var = s2 / n - m * m # variance
if (var < 0) var = 0 # avoid an exception due to round-off error
mean[suffix] = m # store the mean in an array
rmsd[suffix] = sqrt(var)
lowest[suffix] = min
}
prefix=suffix=FILENAME
sub(/_.*/, "", prefix)
sub(/\/[^\/]+$/, "", suffix)
sub(/^.*_/, "", suffix)
s = 0 # sum of $3
s2 = 0 # sum of $3 ** 2
n = 0 # count of samples
min = 0 # lowest value of $3
}
FNR > 1 {
s += $3
s2 += $3 * $3
++n
if ($3 < min) min = $3 # update the lowest value
}
END {
if (n) { # just to avoid division by zero
m = s / n
var = s2 / n - m * m
if (var < 0) var = 0
mean[suffix] = m
rmsd[suffix] = sqrt(var)
lowest[suffix] = min
}
out = rescore "/" prefix ".csv"
print prefix ":", "dG(mean)", "dG(rmsd)", "dG(lowest)" > out
for (i in mean)
printf "%s: %.2f %.2f %.2f\n", i, mean[i], rmsd[i], lowest[i] > out
}' file_*.csv
I've assumed all dG values are negative. If there is any chance the
value is greater than zero, modify the line min = 0 which initializes
the variable to considerably big value (10,000 or whatever).
Please apply your modifications regarding the filenames, if needed.
The suggestions by Ed Morton are also included although the results will be the same.