SJ Cuthbertson

grep -B 3 average stuart.out | grep -v Turnpoint | sed -e 's/cosine/0/g;s/random/1/g;/-/d; s/[[:alpha:][:space:]:]\+/,/g;s/^,8//g; s/,\(([[:digit:]\.]\+)\)//g;s/^,//g' > clean.csv

Yup, I feel that’s quite an impressively long sed argument. It is probably possible to write something more succinct that will have the same effect, but it’d only be even more obfuscated stuart.out is the data collected from me running myself as a subject for my Part II Psychology project: it’s generated by a C program running on DOS 6.2 . I only need three numbers and a word per condition tested – but the program spews out huge amounts of supplementary text, and isn’t totally regular about where newlines get inserted (I had to forced-close the program a number of times). The greps eliminate incomplete session records, and wipe a handful of unnecessary lines from each blockl; the sed statement gets rid of words in between my precious numbers, and makes the textfile look like a csv.

I’m still working on the sed part of the pipe… it needs a final clause to strip certain, but not all, newlines. I need to match every first and second newline out of three, but not the third, in this file (effectively concatenating three lines together onto one line, and repeating for the next three lines). If anybody can lend any hints, please do – not only is my sed-fu weak, but my google-fu seems to be on the blink as well.

Why am I up late doing this? The gutsy upgrade stalled when my net connection broke inexplicably a few hours ago – and I only noticed recently. Only another hour of downloads to go, then the real fun will begin