Simple kevlar utility for reservoir sampling of partitioned reads. The sampling function might be of general interest.

sample.py

#!/usr/bin/env python
import argparse
import kevlar
import random
import sys

def resv_samp(objstream, n=100, stopafter=None, filterfunc=None):
    sample = list()
    for counter, obj in enumerate(objstream):
        if filterfunc:
            discard = filterfunc(obj)
            if discard:
                continue
        if len(sample) < n:
            sample.append(obj)
            continue
        guess = random.randint(1, counter)
        if guess <= n:
            sample[guess-1] = obj
        if stopafter and counter > stopafter:
            break
    return sample

parser = argparse.ArgumentParser()
parser.add_argument('partreads')
parser.add_argument('outfile')
args = parser.parse_args()

reader = kevlar.parse_augmented_fastx(kevlar.open(args.partreads, 'r'))
preader = kevlar.parse_partitioned_reads(reader)
outstream = kevlar.open(args.outfile, 'w')
for partition in resv_samp(preader, filterfunc=lambda p: len(p) > 1000):
    for read in partition:
        kevlar.print_augmented_fastx(read, outstream)