626_privacy/tensorflow_privacy
2
0
Fork 1
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

Add files via upload

Browse source
This commit is contained in:
woodyx218 2020-01-22 10:42:27 +08:00 committed by GitHub
parent 7b72e8a11b
commit bcbb0c9553
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 15 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
tutorials/adult_tutorial.py
View file

@ -43,6 +43,7 @@ flags.DEFINE_integer('epochs', 20, 'Number of epochs')
flags.DEFINE_integer('max_mu', 2, 'GDP upper limit') flags.DEFINE_integer('max_mu', 2, 'GDP upper limit')
flags.DEFINE_string('model_dir', None, 'Model directory') flags.DEFINE_string('model_dir', None, 'Model directory')
sampling_batch = 256
microbatches = 256 microbatches = 256
num_examples = 29305 num_examples = 29305
@ -136,11 +137,11 @@ def main(unused_argv):
shuffle=False) shuffle=False)
# Training loop. # Training loop.
steps_per_epoch = num_examples // microbatches steps_per_epoch = num_examples // sampling_batch
test_accuracy_list = [] test_accuracy_list = []
for epoch in range(1, FLAGS.epochs + 1): for epoch in range(1, FLAGS.epochs + 1):
for step in range(steps_per_epoch): for step in range(steps_per_epoch):
whether = np.random.random_sample(num_examples) > (1-microbatches/num_examples) whether = np.random.random_sample(num_examples) > (1-sampling_batch/num_examples)
subsampling = [i for i in np.arange(num_examples) if whether[i]] subsampling = [i for i in np.arange(num_examples) if whether[i]]
global microbatches global microbatches
microbatches = len(subsampling) microbatches = len(subsampling)
@ -162,8 +163,8 @@ def main(unused_argv):
# Compute the privacy budget expended so far. # Compute the privacy budget expended so far.
if FLAGS.dpsgd: if FLAGS.dpsgd:
eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, 256, 1e-5) eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch, 1e-5)
mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, 256) mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch)
print('For delta=1e-5, the current epsilon is: %.2f' % eps) print('For delta=1e-5, the current epsilon is: %.2f' % eps)
print('For delta=1e-5, the current mu is: %.2f' % mu) print('For delta=1e-5, the current mu is: %.2f' % mu)

10
tutorials/imdb_tutorial.py
View file

@ -43,7 +43,7 @@ flags.DEFINE_integer('epochs', 25, 'Number of epochs')
flags.DEFINE_integer('max_mu', 2, 'GDP upper limit') flags.DEFINE_integer('max_mu', 2, 'GDP upper limit')
flags.DEFINE_string('model_dir', None, 'Model directory') flags.DEFINE_string('model_dir', None, 'Model directory')
sampling_batch = 512
microbatches = 512 microbatches = 512
max_features = 10000 max_features = 10000
@ -137,12 +137,12 @@ def main(unused_argv):
shuffle=False) shuffle=False)
# Training loop. # Training loop.
steps_per_epoch = num_examples // microbatches steps_per_epoch = num_examples // sampling_batch
test_accuracy_list = [] test_accuracy_list = []
for epoch in range(1, FLAGS.epochs + 1): for epoch in range(1, FLAGS.epochs + 1):
for step in range(steps_per_epoch): for step in range(steps_per_epoch):
whether = np.random.random_sample(num_examples) > (1-microbatches/num_examples) whether = np.random.random_sample(num_examples) > (1-sampling_batch/num_examples)
subsampling = [i for i in np.arange(num_examples) if whether[i]] subsampling = [i for i in np.arange(num_examples) if whether[i]]
global microbatches global microbatches
microbatches = len(subsampling) microbatches = len(subsampling)
@ -164,8 +164,8 @@ def main(unused_argv):
# Compute the privacy budget expended so far. # Compute the privacy budget expended so far.
if FLAGS.dpsgd: if FLAGS.dpsgd:
eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, microbatches, 1e-5) eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch, 1e-5)
mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, microbatches) mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch)
print('For delta=1e-5, the current epsilon is: %.2f' % eps) print('For delta=1e-5, the current epsilon is: %.2f' % eps)
print('For delta=1e-5, the current mu is: %.2f' % mu) print('For delta=1e-5, the current mu is: %.2f' % mu)

10
tutorials/movielens_tutorial.py
View file

@ -44,7 +44,7 @@ flags.DEFINE_integer('epochs', 25, 'Number of epochs')
flags.DEFINE_integer('max_mu', 2, 'GDP upper limit') flags.DEFINE_integer('max_mu', 2, 'GDP upper limit')
flags.DEFINE_string('model_dir', None, 'Model directory') flags.DEFINE_string('model_dir', None, 'Model directory')
sampling_batch = 10000
microbatches = 10000 microbatches = 10000
num_examples = 800167 num_examples = 800167
@ -171,11 +171,11 @@ def main(unused_argv):
shuffle=False) shuffle=False)
# Training loop. # Training loop.
steps_per_epoch = num_examples // microbatches steps_per_epoch = num_examples // sampling_batch
test_accuracy_list = [] test_accuracy_list = []
for epoch in range(1, FLAGS.epochs + 1): for epoch in range(1, FLAGS.epochs + 1):
for step in range(steps_per_epoch): for step in range(steps_per_epoch):
whether = np.random.random_sample(num_examples) > (1-microbatches/num_examples) whether = np.random.random_sample(num_examples) > (1-sampling_batch/num_examples)
subsampling = [i for i in np.arange(num_examples) if whether[i]] subsampling = [i for i in np.arange(num_examples) if whether[i]]
global microbatches global microbatches
microbatches = len(subsampling) microbatches = len(subsampling)
@ -197,8 +197,8 @@ def main(unused_argv):
# Compute the privacy budget expended so far. # Compute the privacy budget expended so far.
if FLAGS.dpsgd: if FLAGS.dpsgd:
eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, microbatches, 1e-6) eps = compute_eps_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch, 1e-6)
mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, microbatches) mu = compute_mu_poisson(epoch, FLAGS.noise_multiplier, num_examples, sampling_batch)
print('For delta=1e-6, the current epsilon is: %.2f' % eps) print('For delta=1e-6, the current epsilon is: %.2f' % eps)
print('For delta=1e-6, the current mu is: %.2f' % mu) print('For delta=1e-6, the current mu is: %.2f' % mu)