randomly pick n samples from each word/directory

subsample.py

#!/usr/bin/python3
# randomly pick n samples from each word/directory
# of speech_commands_v0.02/ data set:
# https://download.tensorflow.org/data/speech_commands_v0.02.tgz
# usage: subsample.py -f -n 300 * > ~/list.txt
# tar czf ~/dsc.tgz $(cat ~/list.txt)

import argparse, os, random
from warnings import warn

parser = argparse.ArgumentParser(
    description='randomly pick n samples from each directory',
    formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('-n', '--npd', type=int, default=300, 
    help='pick this many files from each directory')
parser.add_argument('-s', '--randomseed', type=str, default='', 
    help='random seed')
parser.add_argument('-f', '--keepfiles', action='store_true',
    help='also include files on the command line')
parser.add_argument('dirs', nargs='*', help='dir1 dir2 ...')
args = parser.parse_args()

if args.randomseed: random.seed(args.randomseed)

for path in args.dirs:
    if not os.path.isdir(path):
        if args.keepfiles:
            print(path)
        continue
    allfiles = []
    for f in os.listdir(path):
        fp = path+'/'+f
        if os.path.isfile(fp):
            allfiles.append(fp)
    if args.npd:
        chosen = random.sample(allfiles, args.npd) if len(allfiles)>args.npd else allfiles
    for fp in chosen:
        print(fp)