svm2txt

#!/usr/bin/python3 -Wall
# svm2txt: convert svm files to binary training files for fasttext
# usage: svm2txt label < file
# notes: multi labels will be removed in favor of argument label vs non-label
#        data source: http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multilabel.html#rcv1v2%20(topics;%20subsets)
# 20170818 erikt(at)xs4all.nl

import getopt
import re
import sys

COMMAND = sys.argv.pop(0)
USAGE = "usage: "+COMMAND+" label < file"

try: options = getopt.getopt(sys.argv,"",[])
except: sys.exit(USAGE)
if len(options[1]) < 1: sys.exit(USAGE)
label = options[1].pop(0)

patternColon = re.compile(":.*$")
for line in sys.stdin:
    line = line.rstrip()
    fields = line.split()
    if len(fields) > 0:
        labels = fields.pop(0).split(",")
        labelSeen = False
        for i in range(0,len(labels)):
            if labels[i] == label: labelSeen = True
        if labelSeen: print("__label__YES="+label,end="")
        else: print("__label__NO="+label,end="")
        for i in range(0,len(fields)):
            fields[i] = patternColon.sub("",fields[i])
            print(" "+fields[i],end="")
        print("")