def predict_teacher(dataset, nb_teachers):
"""
This is for obtaining the weight from student / teache, don't involve any noise
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param teacher: if teacher is true, then predict with training dataset, else students
:return: out prediction based on cnn
"""
assert input.create_dir_if_needed(FLAGS.train_dir)
train_only = True
test_only = False
# create path to save teacher predict teacher model
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_teacher_clean_votes_label_shift' + str(
FLAGS.lap_scale) + '.npy'
# Load the dataset
if dataset == 'svhn':
test_data, test_labels = input.ld_svhn(test_only, train_only)
elif dataset == 'cifar10':
test_data, test_labels = input.ld_cifar10(test_only, train_only)
elif dataset == 'mnist':
test_data, test_labels = input.ld_mnist(test_only, train_only)
elif dataset == 'adult':
test_data, test_labels = input.ld_adult(test_only, train_only)
else:
print("Check value of dataset flag")
return False
if os.path.exists(filepath):
pred_labels = np.load(filepath)
return pred_labels, test_labels
teachers_preds = ensemble_preds(dataset, nb_teachers, test_data)
# Aggregate teacher predictions to get student training labels
pred_labels = aggregation.noisy_max(FLAGS.nb_teachers, teachers_preds, 0)
utils.save_file(filepath, pred_labels)
# Print accuracy of aggregated labels
ac_ag_labels = metrics.accuracy(pred_labels, test_labels)
print("obtain_weight Accuracy of the aggregated labels: " +
str(ac_ag_labels))
return pred_labels, test_labels
def predict_data(dataset, nb_teachers, teacher=False):
"""
This is for obtaining the weight from student / teache, don't involve any noise
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param teacher: if teacher is true, then predict with training dataset, else students
:return: out prediction based on cnn
"""
assert input.create_dir_if_needed(FLAGS.train_dir)
if teacher:
train_only = True
test_only = False
else:
train_only = False
test_only = True
# Load the dataset
if dataset == 'svhn':
test_data, test_labels = input.ld_svhn(test_only, train_only)
elif dataset == 'cifar10':
test_data, test_labels = input.ld_cifar10(test_only, train_only)
elif dataset == 'mnist':
test_data, test_labels = input.ld_mnist(test_only, train_only)
elif dataset == 'adult':
test_data, test_labels = input.ld_adult(test_only, train_only)
else:
print("Check value of dataset flag")
return False
teachers_preds = ensemble_preds(dataset, nb_teachers, test_data)
# Aggregate teacher predictions to get student training labels
pred_labels = aggregation.noisy_max(FLAGS.nb_teachers, teachers_preds, 0)
# Print accuracy of aggregated labels
ac_ag_labels = metrics.accuracy(pred_labels, test_labels)
print("obtain_weight Accuracy of the aggregated labels: " +
str(ac_ag_labels))
return test_data, pred_labels, test_labels
def prepare_student_data(test_data, nb_teachers, lap_scale):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:Param: lap_scale: scale of the Laplacian noise added for privacy
:return: pairs of (data, labels) to be used for student training and testing
"""
# Compute teacher predictions for student training data
teachers_preds = ensemble_preds(nb_teachers, test_data, 2)
# Aggregate teacher predictions to get student training labels
stdnt_labels = aggregation.noisy_max(teachers_preds, lap_scale)
print('stdnt_labels')
stdnt_labels = keras.utils.to_categorical(stdnt_labels, 2)
print(len(stdnt_labels))
print(stdnt_labels.shape)
# Store unused part of test set for use as a test set after student training
return stdnt_labels
def prepare_student_data(test_data,nb_teachers,epsilon=0.1):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:Param: epsilon: epsilon in (epsilon, delta) differential privacy
:return: pairs of (data, labels) to be used for student training and testing
"""
# Compute teacher predictions for student training data
teachers_preds = ensemble_preds(nb_teachers, test_data, 1)
# Aggregate teacher predictions to get student training labels
stdnt_labels = aggregation.noisy_max(teachers_preds, epsilon=epsilon)
print('stdnt_labels')
#stdnt_labels = tensorflow.keras.utils.to_categorical(stdnt_labels, 1)
print(len(stdnt_labels))
print(stdnt_labels.shape)
# Store unused part of test set for use as a test set after student training
return stdnt_labels
def prepare_student_data(dataset, nb_teachers, save=False, shift_data=None):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model, according to parameters indicated
in flags above.
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param save: if set to True, will dump student training labels predicted by
the ensemble of teachers (with Laplacian noise) as npy files.
It also dumps the clean votes for each class (without noise) and
the labels assigned by teachers
:return: pairs of (data, labels) to be used for student training and testing
"""
stdnt_data = shift_data['data']
test_labels = shift_data['label']
gau_filepath, filepath, filepath_labels = utils.create_path(
FLAGS, dataset, nb_teachers)
if os.path.exists(filepath):
if FLAGS.PATE2 == True:
with open(filepath, 'rb') as f:
clean_votes = np.load(f)
keep_idx, result = gaussian(FLAGS.nb_labels, clean_votes)
precision_true = metrics.accuracy(result,
test_labels[keep_idx])
print('number of idx={} precision_true ={}'.format(
len(keep_idx[0]), precision_true))
return keep_idx, stdnt_data[keep_idx], result
print('*** length of shift_data {} lable length={}********'.format(
len(stdnt_data), len(test_labels)))
# Compute teacher predictions for student training data
teacher_path = 'teacher_pred.npy'
if os.path.exists(teacher_path):
teachers_preds = np.load(teacher_path)
else:
teachers_preds = ensemble_preds(dataset, nb_teachers, stdnt_data)
np.save(teacher_path, teachers_preds)
# Aggregate teacher predictions to get student training labels
if not save:
stdnt_labels = aggregation.noisy_max(FLAGS.nb_labels, teachers_preds,
FLAGS.lap_scale)
else:
# Request clean votes and clean labels as well
stdnt_labels, clean_votes, labels_for_dump = aggregation.noisy_max(
FLAGS.nb_labels,
teachers_preds,
FLAGS.lap_scale,
return_clean_votes=True) #NOLINT(long-line)
# Dump clean_votes array
utils.save_file(filepath, clean_votes)
utils.save_file(filepath_labels, labels_for_dump)
if FLAGS.PATE2 == True:
keep_idx, result = gaussian(FLAGS.nb_labels, clean_votes)
utils.save_file(gau_filepath, result)
ac_ag_labels = metrics.accuracy(result, test_labels[keep_idx])
print(
"number of gaussian student {} Accuracy of the aggregated labels:{} "
.format(len(result), ac_ag_labels))
return keep_idx, stdnt_data[keep_idx], result
def prepare_student_data(dataset, nb_teachers, save=False):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model, according to parameters indicated
in flags above.
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param save: if set to True, will dump student training labels predicted by
the ensemble of teachers (with Laplacian noise) as npy files.
It also dumps the clean votes for each class (without noise) and
the labels assigned by teachers
:return: pairs of (data, labels) to be used for student training and testing
"""
assert input.create_dir_if_needed(FLAGS.train_dir)
# Load the dataset
if dataset == 'svhn':
test_data, test_labels = input.ld_svhn(test_only=True)
elif dataset == 'cifar10':
test_data, test_labels = input.ld_cifar10(test_only=True)
elif dataset == 'mnist':
test_data, test_labels = input.ld_mnist(test_only=True)
else:
print("Check value of dataset flag")
return False
# Make sure there is data leftover to be used as a test set
assert FLAGS.stdnt_share < len(test_data)
# Prepare [unlabeled] student training data (subset of test set)
stdnt_data = test_data[:FLAGS.stdnt_share]
# Compute teacher predictions for student training data
teachers_preds = ensemble_preds(dataset, nb_teachers, stdnt_data)
# Aggregate teacher predictions to get student training labels
if not save:
stdnt_labels = aggregation.noisy_max(teachers_preds, FLAGS.lap_scale)
else:
# Request clean votes and clean labels as well
stdnt_labels, clean_votes, labels_for_dump = aggregation.noisy_max(teachers_preds, FLAGS.lap_scale, return_clean_votes=True) #NOLINT(long-line)
# Prepare filepath for numpy dump of clean votes
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(nb_teachers) + '_student_clean_votes_lap_' + str(FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
# Prepare filepath for numpy dump of clean labels
filepath_labels = FLAGS.data_dir + "/" + str(dataset) + '_' + str(nb_teachers) + '_teachers_labels_lap_' + str(FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
# Dump clean_votes array
with gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, clean_votes)
# Dump labels_for_dump array
with gfile.Open(filepath_labels, mode='w') as file_obj:
np.save(file_obj, labels_for_dump)
# Print accuracy of aggregated labels
ac_ag_labels = metrics.accuracy(stdnt_labels, test_labels[:FLAGS.stdnt_share])
print("Accuracy of the aggregated labels: " + str(ac_ag_labels))
# Store unused part of test set for use as a test set after student training
stdnt_test_data = test_data[FLAGS.stdnt_share:]
stdnt_test_labels = test_labels[FLAGS.stdnt_share:]
if save:
# Prepare filepath for numpy dump of labels produced by noisy aggregation
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(nb_teachers) + '_student_labels_lap_' + str(FLAGS.lap_scale) + '.npy' #NOLINT(long-line)
# Dump student noisy labels array
with gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, stdnt_labels)
return stdnt_data, stdnt_labels, stdnt_test_data, stdnt_test_labels
def prepare_student_data(dataset, nb_teachers, save=False):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model, according to parameters indicated
in flags above.
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param save: if set to True, will dump student training labels predicted by
the ensemble of teachers (with Laplacian noise) as npy files.
It also dumps the clean votes for each class (without noise) and
the labels assigned by teachers
:return: pairs of (data, labels) to be used for student training and testing
"""
assert input.create_dir_if_needed(FLAGS.train_dir)
# Load the dataset
if dataset == 'svhn':
test_data, test_labels = input.ld_svhn(test_only=True)
elif dataset == 'cifar10':
test_data, test_labels = input.ld_cifar10(test_only=True)
elif dataset == 'mnist':
test_data, test_labels = input.ld_mnist(test_only=True)
elif dataset == 'digit':
test_data, test_labels = input.ld_digit_test(test_name=FLAGS.test_name,
num=2000)
else:
print("Check value of dataset flag")
return False
# Make sure there is data leftover to be used as a test set
assert FLAGS.stdnt_share < len(test_data)
# Prepare [unlabeled] student training data (subset of test set)
if (FLAGS.d_stu > -1):
# stdnt_data = []
# for i in range(FLAGS.stdnt_share):
# new_img = transform.resize(skimage.img_as_ubyte(test_data[i].astype(int)),(28,28))
# if FLAGS.d_stu == 3:
# new_img = color.rgb2gray(new_img)
# else:
# new_img = new_img[ :,:, FLAGS.d_stu]
# stdnt_data.append(new_img.reshape(28,28,1).astype(np.float32))
# stdnt_data = np.array(stdnt_data)
trimmed = test_data[:FLAGS.stdnt_share, 2:30, 2:30, :]
# grey scale
if (FLAGS.d_stu == 3):
stdnt_data = 0.2125 * trimmed[:, :, :,
0] + 0.7154 * trimmed[:, :, :,
1] + 0.0721 * trimmed[:, :, :,
2]
else:
stdnt_data = trimmed[:, :, :, FLAGS.d_stu]
stdnt_data = stdnt_data.reshape((-1, 28, 28, 1))
else:
stdnt_data = test_data[:FLAGS.stdnt_share]
# Compute teacher predictions for student training data
teachers_preds = ensemble_preds(dataset, nb_teachers, stdnt_data)
# Aggregate teacher predictions to get student training labels
if not save:
stdnt_labels = aggregation.noisy_max(teachers_preds, FLAGS.lap_scale)
else:
# Request clean votes and clean labels as well
stdnt_labels, clean_votes, labels_for_dump = aggregation.noisy_max(
teachers_preds, FLAGS.lap_scale,
return_clean_votes=True) #NOLINT(long-line)
# Prepare filepath for numpy dump of clean votes
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_student_clean_votes_lap_' + str(
FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
# Prepare filepath for numpy dump of clean labels
filepath_labels = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_teachers_labels_lap_' + str(
FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
# Dump clean_votes array
with tf.gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, clean_votes)
# Dump labels_for_dump array
with tf.gfile.Open(filepath_labels, mode='w') as file_obj:
np.save(file_obj, labels_for_dump)
# Print accuracy of aggregated labels
ac_ag_labels = metrics.accuracy(stdnt_labels,
test_labels[:FLAGS.stdnt_share])
print("Accuracy of the aggregated labels: " + str(ac_ag_labels))
# Store unused part of test set for use as a test set after student training
if FLAGS.dataset_teacher == 'mnist':
test_data, test_labels = input.ld_mnist(test_only=True)
else:
assert 0 == 1, "Non implemented error: dataset_teacher not equals to mnist"
# if FLAGS.d_stu > -1:
# stdnt_test_data = test_data[FLAGS.stdnt_share:, 2:30, 2:30, FLAGS.d_stu : FLAGS.d_stu+1]
# else:
stdnt_test_data = test_data[FLAGS.stdnt_share:]
stdnt_test_labels = test_labels[FLAGS.stdnt_share:]
if save:
# Prepare filepath for numpy dump of labels produced by noisy aggregation
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_student_labels_lap_' + str(
FLAGS.lap_scale) + '.npy' #NOLINT(long-line)
# Dump student noisy labels array
with tf.gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, stdnt_labels)
return stdnt_data, stdnt_labels, stdnt_test_data, stdnt_test_labels
def prepare_student_data(dataset, nb_teachers, save=False, shift_data=None):
"""
Takes a dataset name and the size of the teacher ensemble and prepares
training data for the student model, according to parameters indicated
in flags above.
:param dataset: string corresponding to mnist, cifar10, or svhn
:param nb_teachers: number of teachers (in the ensemble) to learn from
:param save: if set to True, will dump student training labels predicted by
the ensemble of teachers (with Laplacian noise) as npy files.
It also dumps the clean votes for each class (without noise) and
the labels assigned by teachers
:return: pairs of (data, labels) to be used for student training and testing
"""
if dataset == 'svhn':
test_data, test_labels = input.ld_svhn(test_only=True)
elif dataset == 'cifar10':
test_data, test_labels = input.ld_cifar10(test_only=True)
elif dataset == 'mnist':
test_data, test_labels = input.ld_mnist(test_only=True)
elif dataset == 'adult':
test_data, test_labels = input.ld_adult(test_only=True)
else:
print("Check value of dataset flag")
return False
if FLAGS.cov_shift == True:
student_file_name = FLAGS.data + 'PCA_student' + FLAGS.dataset + '.pkl'
f = open(student_file_name, 'rb')
test = pickle.load(f)
test_data = test['data']
test_labels = test['label']
# Prepare [unlabeled] student training data (subset of test set)
stdnt_data = test_data
assert input.create_dir_if_needed(FLAGS.train_dir)
gau_filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_student_votes_sigma1:' + str(
FLAGS.sigma1) + '_sigma2:' + str(
FLAGS.sigma2) + '.npy' # NOLINT(long-line)
# Prepare filepath for numpy dump of clean votes
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_student_clean_votes' + str(
FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
# Prepare filepath for numpy dump of clean labels
filepath_labels = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_teachers_labels_lap_' + str(
FLAGS.lap_scale) + '.npy' # NOLINT(long-line)
"""
if os.path.exists(filepath):
if FLAGS.PATE2 == True:
with open(filepath,'rb')as f:
clean_votes = np.load(f)
keep_idx, result = gaussian(FLAGS.nb_labels, clean_votes)
precision_true = metrics.accuracy(result, test_labels[keep_idx])
print('number of idx={}'.format(len(keep_idx[0])))
return keep_idx, stdnt_data[keep_idx], result
"""
# Load the dataset
# Make sure there is data leftover to be used as a test set
assert FLAGS.stdnt_share < len(test_data)
if shift_data is not None:
#no noise
# replace original student data with shift data
stdnt_data = shift_data['data']
test_labels = shift_data['label']
print('*** length of shift_data {} lable length={}********'.format(
len(stdnt_data), len(test_labels)))
# Compute teacher predictions for student training data
teachers_preds = ensemble_preds(dataset, nb_teachers, stdnt_data)
# Aggregate teacher predictions to get student training labels
if not save:
stdnt_labels = aggregation.noisy_max(teachers_preds, FLAGS.lap_scale)
else:
# Request clean votes and clean labels as well
stdnt_labels, clean_votes, labels_for_dump = aggregation.noisy_max(
FLAGS.nb_labels,
teachers_preds,
FLAGS.lap_scale,
return_clean_votes=True) #NOLINT(long-line)
if FLAGS.PATE2 == True:
keep_idx, result = gaussian(FLAGS.nb_labels, clean_votes)
# Dump clean_votes array
with tf.gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, clean_votes)
# Dump labels_for_dump array
with tf.gfile.Open(filepath_labels, mode='w') as file_obj:
np.save(file_obj, labels_for_dump)
# Print accuracy of aggregated labels
if FLAGS.PATE2 == True:
with tf.gfile.Open(gau_filepath, mode='w') as file_obj:
np.save(file_obj, result)
ac_ag_labels = metrics.accuracy(result, test_labels[keep_idx])
print(
"number of gaussian student {} Accuracy of the aggregated labels:{} "
.format(len(result), ac_ag_labels))
return keep_idx, stdnt_data[keep_idx], result
else:
ac_ag_labels = metrics.accuracy(stdnt_labels, test_labels)
print("Accuracy of the aggregated labels: " + str(ac_ag_labels))
if save:
# Prepare filepath for numpy dump of labels produced by noisy aggregation
filepath = FLAGS.data_dir + "/" + str(dataset) + '_' + str(
nb_teachers) + '_student_labels_lap_' + str(
FLAGS.lap_scale) + '.npy' #NOLINT(long-line)
# Dump student noisy labels array
with tf.gfile.Open(filepath, mode='w') as file_obj:
np.save(file_obj, stdnt_labels)
return stdnt_data, stdnt_labels
