def compare_alias_and_inverse_cdf():
from project_utils import inverse_cdf_sampler
import time
from utils import qqplot
from ticktock import tic, toc
G = 5000000
q = 50
num_samples = 1000000
ps = [random.random() * 50 / float(G) for i in range(G)]
tic("inverse sampler")
inv_sampler = inverse_cdf_sampler(ps)
toc()
tic("inverse sampling")
inv_samples = [inv_sampler() for i in range(num_samples)]
toc()
tic("alias sampler")
al_sampler = alias_sampler(ps)
toc()
tic("alias sampling")
al_samples = [al_sampler() for i in range(num_samples)]
toc()
return inv_samples, al_samples
def alias_sampler(_ps):
tic("setup")
ps = np.copy(_ps)
ps /= np.sum(ps)
n = len(ps)
alias = np.zeros(n)
prob = np.zeros(n)
ps *= n
toc()
# could vectorize this?...
tic("initializing small and large")
# for i in xrange(n):
# if ps[i] < 1:
# small.append(i)
# else:
# large.append(i)
small = ps[ps < 1]
large = ps[ps >= 1]
toc()
tic("while small and large")
sidx = 0
lidx = 0
sn = len(small)
ln = len(large)
# while np.any(small) and np.any(large):
while sidx < len(small) and lidx < len(large):
# l = small.pop()
# g = large.pop()
s = small[sidx]
l = large[lidx]
pl = ps[l]
prob[l] = pl
alias[l] = g
ps[g] += pl - 1
if ps[g] < 1:
# small.append(g)
np.append(small, g)
else:
# large.append(g)
np.append(large, g)
sidx += 1
lidx += 1
toc()
tic("while large")
while large:
g = large.pop()
prob[g] = 1
toc()
tic("while small")
while small:
l = small.pop()
prob[l] = 1
toc()
def sampler():
i = random.randrange(n)
if random.random() < prob[i]:
return i
else:
return alias[i]
return sampler
