If you're just looking for a python version of R's sample function, try this:
import collections
import random
import bisect
def sample(xs, sample_size = None, replace=False, sample_probabilities = None):
"""Mimics the functionality of http://statistics.ats.ucla.edu/stat/r/library/bootstrap.htm sample()"""
if not isinstance(xs, collections.Iterable):
xs = range(xs)
if not sample_size:
sample_size = len(xs)
if not sample_probabilities:
if replace:
return [random.choice(xs) for _ in range(sample_size)]
else:
return random.sample(xs, sample_size)
else:
if replace:
total, cdf = 0, []
for x, p in zip(xs, sample_probabilities):
total += p
cdf.append(total)
return [ xs[ bisect.bisect(cdf, random.uniform(0, total)) ]
for _ in range(sample_size) ]
else:
assert len(sample_probabilities) == len(xs)
xps = list(zip(xs, sample_probabilities))
total = sum(sample_probabilities)
result = []
for _ in range(sample_size):
# choose an item based on weights, and remove it from future iterations.
# this is slow (N^2), a tree structure for xps would be better (NlogN)
target = random.uniform(0, total)
current_total = 0
for index, (x,p) in enumerate(xps):
current_total += p
if current_total > target:
xps.pop(index)
result.append(x)
total -= p
break
return result
