def test_rvs(self):
vals = stats.dlaplace.rvs(1.5, size=(2, 50))
assert numpy.shape(vals) == (2, 50)
assert vals.dtype.char in typecodes["AllInteger"]
val = stats.dlaplace.rvs(1.5)
assert isinstance(val, int)
val = stats.dlaplace(1.5).rvs(3)
assert isinstance(val, numpy.ndarray)
assert val.dtype.char in typecodes["AllInteger"]
def test_rvs(self):
vals = stats.dlaplace.rvs(1.5 , size=(2, 50))
assert_(numpy.shape(vals) == (2, 50))
assert_(vals.dtype.char in typecodes['AllInteger'])
val = stats.dlaplace.rvs(1.5)
assert_(isinstance(val, int))
val = stats.dlaplace(1.5).rvs(3)
assert_(isinstance(val, numpy.ndarray))
assert_(val.dtype.char in typecodes['AllInteger'])
def test_stats(self):
# compare the explicit formulas w/ direct summation using pmf
a = 1.
dl = stats.dlaplace(a)
m, v, s, k = dl.stats('mvsk')
N = 37
xx = np.arange(-N, N+1)
pp = dl.pmf(xx)
m2, m4 = np.sum(pp*xx**2), np.sum(pp*xx**4)
assert_equal((m, s), (0,0))
assert_allclose((v, k), (m2, m4/m2**2 - 3.), atol=1e-14, rtol=1e-8)
def __init__(self,
parameter: float,
sensitivity: int = 1) -> None:
"""Initializes the privacy loss of the discrete Laplace mechanism.
Args:
parameter: the parameter of the discrete Laplace distribution.
sensitivity: the sensitivity of function f. (i.e. the maximum absolute
change in f when an input to a single user changes.)
"""
if parameter <= 0:
raise ValueError(f'Parameter is not a positive real number: {parameter}')
if not isinstance(sensitivity, int):
raise ValueError(f'Sensitivity is not an integer : {sensitivity}')
self.sensitivity = sensitivity
self._parameter = parameter
self._discrete_laplace_random_variable = stats.dlaplace(parameter)
super(DiscreteLaplacePrivacyLoss, self).__init__(sensitivity, True)
def test_laplace(self):
from scipy.stats import dlaplace
import matplotlib.pyplot as plt
import numpy as np
fig, ax = plt.subplots(1, 1)
a = 0.8
mean, var, skew, kurt = dlaplace.stats(a, moments='mvsk')
x = np.arange(dlaplace.ppf(0.01, a), dlaplace.ppf(0.99, a))
ax.plot(x, dlaplace.pmf(x, a), 'bo', ms=8, label='dlaplace pmf')
ax.vlines(x, 0, dlaplace.pmf(x, a), colors='b', lw=5, alpha=0.5)
rv = dlaplace(a)
ax.vlines(x,
0,
rv.pmf(x),
colors='k',
linestyles='-',
lw=1,
label='frozen pmf')
ax.legend(loc='best', frameon=False)
self.assertEqual(str(ax), "AxesSubplot(0.125,0.11;0.775x0.77)")
})
def build_layer_fcr(base_rv, decay_rate):
return lambda prev, layer_num: (base_rv
if prev is None else MixtureDistribution([
poisson(prev * decay_rate, loc=2),
DeltaDistribution(loc=prev),
poisson(prev / decay_rate, loc=2),
]))
NUM_GRAPH_LAYERS_RV = geom(0.5)
DENSE_LAYERS_RV = LayerDistribution(
num_layer_distribution=geom(0.5, loc=-1),
layer_size_distribution_fn=build_layer_fcr(dlaplace(1e-1, loc=150), 0.5),
)
def constant_int_fcr(choices):
n = len(choices)
return lambda _1, _2: CategoricalRV(choices)
def filter_K_fcr(base_rv, delta_rate):
return lambda prev, _: (base_rv if prev is None else MixtureDistribution([
poisson(prev * delta_rate, loc=2),
DeltaDistribution(loc=prev),
poisson(prev / delta_rate, loc=2),
]))
a = 0.8
mean, var, skew, kurt = dlaplace.stats(a, moments='mvsk')
# Display the probability mass function (``pmf``):
x = np.arange(dlaplace.ppf(0.01, a), dlaplace.ppf(0.99, a))
ax.plot(x, dlaplace.pmf(x, a), 'bo', ms=8, label='dlaplace pmf')
ax.vlines(x, 0, dlaplace.pmf(x, a), colors='b', lw=5, alpha=0.5)
# Alternatively, the distribution object can be called (as a function)
# to fix the shape and location. This returns a "frozen" RV object holding
# the given parameters fixed.
# Freeze the distribution and display the frozen ``pmf``:
rv = dlaplace(a)
ax.vlines(x,
0,
rv.pmf(x),
colors='k',
linestyles='-',
lw=1,
label='frozen pmf')
ax.legend(loc='best', frameon=False)
plt.show()
# Check accuracy of ``cdf`` and ``ppf``:
prob = dlaplace.cdf(x, a)
np.allclose(x, dlaplace.ppf(prob, a))
# True
def test_stats2(self):
a = np.log(2.)
dl = stats.dlaplace(a)
m, v, s, k = dl.stats('mvsk')
assert_equal((m, s), (0.,0.))
assert_allclose((v, k), (4., 3.25))
