def test_exponential_mechanism_sparse(mechanism, test_pick=True):
if test_pick:
test_pick_out_of_sub_group()
d = np.random.normal(0, 1, 1000)
r = xrange(-2**30, 2**30)
ri = range(-4, 4)
# print ri
result = bdp.sparse_domain(mechanism, d, r, ri, ql.quality_minmax, 0.1)
exact_median = np.median(d)
print "true median = %f" % exact_median
print "and its quality is : %d" % ql.quality_minmax(d, exact_median)
print "result = %f" % result
print "and its quality is : %d" % ql.quality_minmax(d, result)
def test_exponential_mechanism_sparse(mechanism, test_pick=True):
if test_pick:
test_pick_out_of_sub_group()
d = np.random.normal(0, 1, 1000)
r = xrange(-2**30, 2**30)
ri = range(-4, 4)
# print ri
result = bdp.sparse_domain(mechanism, d, r, ri, ql.quality_minmax, 0.1)
exact_median = np.median(d)
print "true median = %f" % exact_median
print "and its quality is : %d" % ql.quality_minmax(d, exact_median)
print "result = %f" % result
print "and its quality is : %d" % ql.quality_minmax(d, result)
def check(t, alpha, eps, delta, beta, samples_size=0, use_exponential=True):
"""
run flat_concave on random data using the given parameters
:param t:
:param alpha:
:param eps:
:param delta:
:param beta:
:param samples_size:
:param use_exponential:
:return:
"""
range_end = 2**t
if samples_size == 0:
samples_size = int(
src.bounds.step6_n2_bound(range_end, eps, alpha, beta))
data_center = np.random.uniform(range_end / 3, range_end / 3 * 2)
# data = src.examples.get_random_data(samples_size, distribution_type='splitted', pivot=data_center)
# data = src.examples.get_random_data(samples_size, 'bimodal')
data = src.examples.get_random_data(samples_size, pivot=data_center)
data = sorted(filter(lambda x: 0 <= x <= range_end, data))
maximum_quality = min_max_maximum_quality(data, 0, range_end)
quality_result_lower_bound = maximum_quality * (1 - alpha)
try:
result = src.flat_concave.evaluate(data, range_end, quality_minmax,
maximum_quality, alpha, eps, delta,
min_max_intervals_bounding,
min_max_maximum_quality,
use_exponential)
result_quality = quality_minmax(data, result)
except ValueError:
# result = -1
result_quality = -1
return result_quality != -1, result_quality >= quality_result_lower_bound, result_quality / float(
maximum_quality)
def check(t, alpha, eps, delta, beta, samples_size=0, use_exponential=True):
"""
run flat_concave on random data using the given parameters
:param t:
:param alpha:
:param eps:
:param delta:
:param beta:
:param samples_size:
:param use_exponential:
:return:
"""
range_end = 2**t
if samples_size == 0:
samples_size = int(src.bounds.step6_n2_bound(range_end, eps, alpha, beta))
data_center = np.random.uniform(range_end/3, range_end/3*2)
# data = src.examples.get_random_data(samples_size, distribution_type='splitted', pivot=data_center)
# data = src.examples.get_random_data(samples_size, 'bimodal')
data = src.examples.get_random_data(samples_size, pivot=data_center)
data = sorted(filter(lambda x: 0 <= x <= range_end, data))
maximum_quality = min_max_maximum_quality(data, 0, range_end)
quality_result_lower_bound = maximum_quality * (1-alpha)
try:
result = src.flat_concave.evaluate(data, range_end, quality_minmax, maximum_quality, alpha, eps, delta,
min_max_intervals_bounding, min_max_maximum_quality,
use_exponential)
result_quality = quality_minmax(data, result)
except ValueError:
# result = -1
result_quality = -1
return result_quality != -1, result_quality >= quality_result_lower_bound, result_quality/float(maximum_quality)
