ruby/tensorflow_privacy
1
0
Fork 0
forked from 626_privacy/tensorflow_privacy
Code Pull requests Activity

resolve space issues

Browse source
This commit is contained in:
Yuqing 2021-05-11 00:19:52 -07:00
parent 09270afed6
commit 9d13376707
1 changed files with 7 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
tensorflow_privacy/privacy/analysis/rdp_accountant.py
View file

@ -273,7 +273,7 @@ def _compute_eps(orders, rdp, delta):
# This bound is not numerically stable as alpha->1. # This bound is not numerically stable as alpha->1.
# Thus we have a min value of alpha. # Thus we have a min value of alpha.
# The bound is also not useful for small alpha, so doesn't matter. # The bound is also not useful for small alpha, so doesn't matter.
eps = r + math.log1p(-1/a) - math.log(delta * a) / (a - 1) eps = r + math.log1p(-1 / a) - math.log(delta * a) / (a - 1)
else: else:
# In this case we can't do anything. E.g., asking for delta = 0. # In this case we can't do anything. E.g., asking for delta = 0.
eps = np.inf eps = np.inf
@ -330,7 +330,7 @@ def _get_forward_diffs(fun, n):
func_vec[i] = fun(1.0 * (i - 1)) func_vec[i] = fun(1.0 * (i - 1))
for i in range(0, n + 2, 1): for i in range(0, n + 2, 1):
# Diff in log scale # Diff in log scale
_stable_inplace_diff_in_log(func_vec, signs_func_vec, n=n + 2 - i) _stable_inplace_diff_in_log(func_vec, signs_func_vec, n = n + 2 - i)
deltas[i] = func_vec[0] deltas[i] = func_vec[0]
signs_deltas[i] = signs_func_vec[0] signs_deltas[i] = signs_func_vec[0]
return deltas, signs_deltas return deltas, signs_deltas
@ -462,11 +462,11 @@ def _compute_rdp_sample_without_replacement_int(q, sigma, alpha):
def cgf(x): def cgf(x):
# Return rdp(x+1)*x, the rdp of Gaussian mechanism is alpha/(2*sigma**2) # Return rdp(x+1)*x, the rdp of Gaussian mechanism is alpha/(2*sigma**2)
return x*1.0*(x+1)/(2.0*sigma**2) return x * 1.0 * (x + 1) / (2.0 * sigma**2)
def func(x): def func(x):
# Return the rdp of Gaussian mechanism # Return the rdp of Gaussian mechanism
return 1.0*(x)/(2.0*sigma**2) return 1.0 * x / (2.0 * sigma**2)
@ -485,7 +485,7 @@ def _compute_rdp_sample_without_replacement_int(q, sigma, alpha):
if i == 2: if i == 2:
s = 2 * np.log(q) + _log_comb(alpha, 2) + np.minimum(np.log(4) + log_f2m1, func(2.0) + np.log(2)) s = 2 * np.log(q) + _log_comb(alpha, 2) + np.minimum(np.log(4) + log_f2m1, func(2.0) + np.log(2))
elif i > 2: elif i > 2:
delta_lo = deltas[int(2*np.floor(i/2.0))-1] delta_lo = deltas[int(2 * np.floor( i / 2.0))-1]
delta_hi = deltas[int(2 * np.ceil(i / 2.0)) - 1] delta_hi = deltas[int(2 * np.ceil(i / 2.0)) - 1]
s = np.log(4) + 0.5 * (delta_lo + delta_hi) s = np.log(4) + 0.5 * (delta_lo + delta_hi)
s = np.minimum(s, np.log(2) + cgf(i - 1)) s = np.minimum(s, np.log(2) + cgf(i - 1))
@ -496,10 +496,10 @@ def _compute_rdp_sample_without_replacement_int(q, sigma, alpha):
# Compute the bound with stirling approximation. Everything is O(x) now. # Compute the bound with stirling approximation. Everything is O(x) now.
for i in range(2, alpha + 1): for i in range(2, alpha + 1):
if i == 2: if i == 2:
s = 2 * np.log(q) + _log_comb(alpha,2) + np.minimum( s = 2 * np.log(q) + _log_comb(alpha, 2) + np.minimum(
np.log(4) + log_f2m1, func(2.0) + np.log(2)) np.log(4) + log_f2m1, func(2.0) + np.log(2))
else: else:
s = np.log(2) + cgf(i-1) + i*np.log(q) + _log_comb(alpha, i) s = np.log(2) + cgf(i - 1) + i*np.log(q) + _log_comb(alpha, i)
log_a = _log_add(log_a, s) log_a = _log_add(log_a, s)
return log_a return log_a