def two_sample(A, B, limit = 10000, progress_bar = None):
""" Conducts a permutation test on the input data, transformed by fun. """
# apply transformation to input data (e.g. signed-rank for WMW)
data = np.concatenate((A, B))
stat_ref = np.sum(A)
# count permutation test statistics <=, >=, or ||>=|| than reference stat
counts = np.array([0,0,0]) # (lesser, greater)
total_perms = bincoef(len(data), len(A))
if not limit or total_perms < limit :
limit = None
comb_function = itertools.combinations
else:
comb_function = random_combination
if progress_bar:
progress_bar.max = limit or total_perms
progress_bar.label = "of %d permutations" % progress_bar.max
for binary_row in binary_combinations(len(data), len(A), comb_function=comb_function, limit=limit):
#print binary_row
stat_this = np.sum(np.array(data)*binary_row)
counts = counts + stat_compare(stat_ref,stat_this)
# if progress_bar:
# progress_bar.advance()
# return p-values for lower, upper, and two-tail tests (FP number)
n_comb = np.multiply.reduce(np.array(range(len(data)-len(A)+1,len(data)+1)))\
/ np.multiply.reduce(np.array(range(1,len(A)+1)))
n_comb = limit or total_perms
counts[2] = min(2*counts[0:2].min(),n_comb) # hack to define p.twotail as 2*smaller of 1 tail p's
return counts / float(n_comb)
def random_combination(iterable, r, limit=None):
""" Random selection from itertools.combinations(iterable, r). """
pool = tuple(iterable)
n = len(pool)
limiter = 0
comb = bincoef(len(iterable), r)
print comb
comb_indices = random.sample(xrange(comb), limit)
while True if limit == None else limiter < limit:
print comb_indices[limiter]
perm = get_nth_perm(pool, comb_indices[limiter])
subpool = sorted(perm[:r])
indices = sorted(random.sample(xrange(n), r))
print tuple(pool[i] for i in indices), subpool, perm
limiter = limiter + 1
yield tuple(pool[i] for i in indices)
def random_combination(iterable, r, limit=None):
""" Random selection from itertools.combinations(iterable, r). """
pool = tuple(iterable)
n = len(pool)
limiter = 0
comb = bincoef(len(iterable), r)
print comb
comb_indices = random.sample(xrange(comb), limit)
while True if limit == None else limiter < limit:
print comb_indices[limiter]
perm = get_nth_perm(pool, comb_indices[limiter])
subpool = sorted(perm[:r])
indices = sorted(random.sample(xrange(n), r))
print tuple(pool[i] for i in indices), subpool, perm
limiter = limiter + 1
yield tuple(pool[i] for i in indices)
def two_sample(A, B, limit=10000, progress_bar=None):
""" Conducts a permutation test on the input data, transformed by fun. """
# apply transformation to input data (e.g. signed-rank for WMW)
data = np.concatenate((A, B))
stat_ref = np.sum(A)
# count permutation test statistics <=, >=, or ||>=|| than reference stat
counts = np.array([0, 0, 0]) # (lesser, greater)
total_perms = bincoef(len(data), len(A))
if not limit or total_perms < limit:
limit = None
comb_function = itertools.combinations
else:
comb_function = random_combination
if progress_bar:
progress_bar.max = limit or total_perms
progress_bar.label = "of %d permutations" % progress_bar.max
for binary_row in binary_combinations(len(data),
len(A),
comb_function=comb_function,
limit=limit):
#print binary_row
stat_this = np.sum(np.array(data) * binary_row)
counts = counts + stat_compare(stat_ref, stat_this)
# if progress_bar:
# progress_bar.advance()
# return p-values for lower, upper, and two-tail tests (FP number)
n_comb = np.multiply.reduce(np.array(range(len(data)-len(A)+1,len(data)+1)))\
/ np.multiply.reduce(np.array(range(1,len(A)+1)))
n_comb = limit or total_perms
counts[2] = min(
2 * counts[0:2].min(),
n_comb) # hack to define p.twotail as 2*smaller of 1 tail p's
return counts / float(n_comb)
