def __call__(self, n, t, y_t):
if self.min_stop_check(t):
return False
y = y_t
not_y = t - y
# The amount for winner under alternative
p1N = int(n * self.p)
# The amount for winner under null
p0N = int(n * self.p_0)
# The amount for winner is greater than alternative
if y >= p1N:
return True
# The amount for loser is greater than votes in null hypothesis for loser
elif not_y >= n - p0N:
return False
# log(top)
log_prob_p1 = hypergeom_pmf(y, n, p1N, t, log=True)
# log(bot)
log_prob_p0 = hypergeom_pmf(y, n, p0N, t, log=True)
inf_p0 = isinf(log_prob_p0)
inf_p1 = isinf(log_prob_p1)
# Both return null, means the chance is in between
if inf_p0 and inf_p1:
#
if y > p0N:
return True
return False
# Chance for null is zero
elif inf_p0:
return True
# Chance for alternative is zero
elif inf_p1:
return False
# total log(top) - log(bottom)
ratio = log_prob_p1 - log_prob_p0
# print(n, t, y_t, log_prob_p0, log_prob_p1, ratio, self.critical_value)
return ratio >= self.critical_value
def single_node_update(rejection_fn, n, t, y_t, p_t, step, p, replacement,
*args, **kwargs):
rejection_dict = defaultdict(float)
q = OrderedDict()
# Take the floor for winner's share
w = floor(n * p)
t_next = t + step
n_remain = n - t
w_remain = w - y_t
# All possible generated sa for next batch
for i in range(step + 1):
y_t_next = y_t + i
if not replacement:
# If no replacement sample from hypergeometric distribution
p_next = hypergeom_pmf(i, n_remain, w_remain, step)
else:
# Else use binomial compute probability
p_next = binom_pmf(i, step, p)
# Is this state rejected?
reject = rejection_fn(n, t_next, y_t_next, *args, **kwargs)
# Compose the node
node = (t_next, y_t_next, p_next * p_t)
if isnull(node[2]):
node = (t_next, y_t_next, 0)
# Don't add to queue or dict if there is no chance for it
if node[2] == 0:
continue
# if null is rejected, put it in the risk dict
if reject:
rejection_dict[node[:2]] += node[2]
else:
q.append(node[:2], node[2])
return rejection_dict, q
def audit_process_simulation_serial(rejection_fn,
election: Election,
progression=False,
*args,
**kwargs):
m = election.m
n = election.n
step = election.step
replacement = election.replacement
p = election.p
if m == -1:
m = n + 1
w = floor(n * p)
progression_bar = null_bar
if progression:
progression_bar = SimpleProgressionBar(m)
rejection_dict = defaultdict(float)
# first element: t,
# second element: y_t, (observe every step time)
# third element: probability going to this state
source = (0, 0, 1)
q = OrderedDict()
q.append(source[:2], source[2])
# While q is not empty
while len(q):
# get next node to explore
key, value = q.peek()
# Update progression
progression_bar(key[0])
# If sampled to the max number already, break (max number exclusive)
if isinstance(m, int) and key[0] >= m:
break
if replacement and key[0] > n:
break
# Remove the value
q.popitem(last=False)
t, y_t, p_t = *key, value
console_logger.debug(" poped: {}".format((t, y_t, p_t)))
n_remain = n - t
w_remain = w - y_t
t_next = t + step
reject = False
# All possible generated sa for next batch
for i in range(0, step + 1):
y_t_next = y_t + i
if not replacement:
n_remain_next = n - t_next
w_remain_next = w - y_t_next
l_remain_next = n_remain_next - w_remain_next
if n_remain_next < 0 or w_remain_next < 0 or l_remain_next < 0:
continue
if y_t_next < 0:
break
if not replacement:
# If no replacement sample from hypergeometric distribution
p_next = hypergeom_pmf(i, n_remain, w_remain, step)
else:
# Else use binomial compute probability
p_next = binom_pmf(i, step, p)
# Update reject if it's False, don't need to compute reject if it's True already (as is tested sequentially)
if not reject:
# Remainder: Don't use is to compare True False
reject = rejection_fn(n, t_next, y_t_next, *args, **kwargs)
# Compose the node
node = (t_next, y_t_next, p_next * p_t)
if isnull(p_next * p_t):
node = (t_next, y_t_next, 0)
# if null is rejected, put it in the risk dict
if reject:
rejection_dict[node[:2]] += node[2]
else:
q.append(node[:2], node[2])
# The information in this is enough to determine the result
return rejection_dict
from utility.program_utility import CachedMethod, string_to_num