hummingbird
/
mindarmour
Not watched
2
0
Fork 0
Code
Releases 17 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain
Browse Source

Fix API Issues

tags/v1.8.0
shu-kun-zhang 3 years ago
parent
49d6cf1feb
commit
8cc2a94549
12 changed files with 285 additions and 88 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +0
    -3
      mindarmour/adv_robustness/defenses/adversarial_defense.py
  2. +3
    -34
      mindarmour/adv_robustness/detectors/black/similarity_detector.py
  3. +1
    -1
      mindarmour/fuzz_testing/fuzzing.py
  4. +218
    -8
      mindarmour/fuzz_testing/model_coverage_metrics.py
  5. +0
    -9
      mindarmour/privacy/diff_privacy/monitor/monitor.py
  6. +4
    -3
      mindarmour/privacy/evaluation/attacker.py
  7. +0
    -1
      mindarmour/privacy/evaluation/inversion_attack.py
  8. +4
    -5
      mindarmour/privacy/evaluation/membership_inference.py
  9. +0
    -7
      mindarmour/reliability/concept_drift/concept_drift_check_time_series.py
  10. +1
    -2
      mindarmour/reliability/model_fault_injection/fault_injection.py
  11. +49
    -10
      mindarmour/utils/util.py
  12. +5
    -5
      tests/ut/python/privacy/evaluation/test_attacker.py

+ 0
- 3
mindarmour/adv_robustness/defenses/adversarial_defense.py View File

@@ -79,9 +79,6 @@ class AdversarialDefense(Defense):


Returns: Returns:
numpy.ndarray, loss of defense operation. numpy.ndarray, loss of defense operation.

Examples:
>>> adv_defense.defense(inputs, labels)
""" """
inputs, labels = check_pair_numpy_param('inputs', inputs, 'labels', inputs, labels = check_pair_numpy_param('inputs', inputs, 'labels',
labels) labels)


+ 3
- 34
mindarmour/adv_robustness/detectors/black/similarity_detector.py View File

@@ -98,6 +98,9 @@ class SimilarityDetector(Detector):
>>> detector.set_threshold(num_nearest_neighbors[-1], thresholds[-1]) >>> detector.set_threshold(num_nearest_neighbors[-1], thresholds[-1])
>>> detector.detect(benign_queries) >>> detector.detect(benign_queries)
>>> detections = detector.get_detection_interval() >>> detections = detector.get_detection_interval()
>>> detector.detect_diff()
>>> detected_queries = detector.get_detected_queries()
>>> detector.transform(x_train)
""" """


def __init__(self, trans_model, max_k_neighbor=1000, chunk_size=1000, def __init__(self, trans_model, max_k_neighbor=1000, chunk_size=1000,
@@ -137,13 +140,6 @@ class SimilarityDetector(Detector):
Raises: Raises:
ValueError: The number of training data is less than ValueError: The number of training data is less than
max_k_neighbor! max_k_neighbor!

Examples:
>>> x_train = np.random.rand(10, 32, 32, 3).astype(np.float32)
>>> perm = np.random.permutation(x_train.shape[0])
>>> benign_queries = x_train[perm[:10], :, :, :]
>>> num_nearest_neighbors, thresholds = detector.fit(inputs=x_train)
>>> detector.set_threshold(num_nearest_neighbors[-1], thresholds[-1])
""" """
data = check_numpy_param('inputs', inputs) data = check_numpy_param('inputs', inputs)
data_len = data.shape[0] data_len = data.shape[0]
@@ -192,14 +188,6 @@ class SimilarityDetector(Detector):
Raises: Raises:
ValueError: The parameters of threshold or num_of_neighbors is ValueError: The parameters of threshold or num_of_neighbors is
not available. not available.

Examples:
>>> x_train = np.random.rand(10, 32, 32, 3).astype(np.float32)
>>> perm = np.random.permutation(x_train.shape[0])
>>> benign_queries = x_train[perm[:10], :, :, :]
>>> num_nearest_neighbors, thresholds = detector.fit(inputs=x_train)
>>> detector.set_threshold(num_nearest_neighbors[-1], thresholds[-1])
>>> detector.detect(benign_queries)
""" """
if self._threshold is None or self._num_of_neighbors is None: if self._threshold is None or self._num_of_neighbors is None:
msg = 'Explicit detection threshold and number of nearest ' \ msg = 'Explicit detection threshold and number of nearest ' \
@@ -247,9 +235,6 @@ class SimilarityDetector(Detector):
def clear_buffer(self): def clear_buffer(self):
""" """
Clear the buffer memory. Clear the buffer memory.

Examples:
>>> detector.detect(benign_queries)
""" """
while self._buffer: while self._buffer:
self._buffer.pop() self._buffer.pop()
@@ -261,10 +246,6 @@ class SimilarityDetector(Detector):
Args: Args:
num_of_neighbors (int): Number of the nearest neighbors. num_of_neighbors (int): Number of the nearest neighbors.
threshold (float): Detection threshold. threshold (float): Detection threshold.

Examples:
>>> num_nearest_neighbors, thresholds = detector.fit(inputs=x_train)
>>> detector.set_threshold(num_nearest_neighbors[-1], thresholds[-1])
""" """
self._num_of_neighbors = check_int_positive('num_of_neighbors', self._num_of_neighbors = check_int_positive('num_of_neighbors',
num_of_neighbors) num_of_neighbors)
@@ -276,9 +257,6 @@ class SimilarityDetector(Detector):


Returns: Returns:
list[int], number of queries between adjacent detections. list[int], number of queries between adjacent detections.

Examples:
>>> detector.get_detection_interval()
""" """
detected_queries = self._detected_queries detected_queries = self._detected_queries
interval = [] interval = []
@@ -292,9 +270,6 @@ class SimilarityDetector(Detector):


Returns: Returns:
list[int], sequence number of detected malicious queries. list[int], sequence number of detected malicious queries.

Examples:
>>> detector.get_detected_queries()
""" """
detected_queries = self._detected_queries detected_queries = self._detected_queries
return detected_queries return detected_queries
@@ -311,9 +286,6 @@ class SimilarityDetector(Detector):
Raises: Raises:
NotImplementedError: This function is not available NotImplementedError: This function is not available
in class `SimilarityDetector`. in class `SimilarityDetector`.

Examples:
>>> detector.detect_diff()
""" """
msg = 'The function detect_diff() is not available in the class ' \ msg = 'The function detect_diff() is not available in the class ' \
'`SimilarityDetector`.' '`SimilarityDetector`.'
@@ -329,9 +301,6 @@ class SimilarityDetector(Detector):


Raises: Raises:
NotImplementedError: This function is not available in class `SimilarityDetector`. NotImplementedError: This function is not available in class `SimilarityDetector`.

Examples:
>>> detector.transform(x_train)
""" """
msg = 'The function transform() is not available in the class `SimilarityDetector`.' msg = 'The function transform() is not available in the class `SimilarityDetector`.'
LOGGER.error(TAG, msg) LOGGER.error(TAG, msg)


+ 1
- 1
mindarmour/fuzz_testing/fuzzing.py View File

@@ -124,7 +124,7 @@ class Fuzzer:
... self.reshape = P.Reshape() ... self.reshape = P.Reshape()
... self.summary = TensorSummary() ... self.summary = TensorSummary()
... ...
... def construct(self, x):
... def construct(self, x):
... x = self.conv1(x) ... x = self.conv1(x)
... x = self.relu(x) ... x = self.relu(x)
... self.summary('conv1', x) ... self.summary('conv1', x)


+ 218
- 8
mindarmour/fuzz_testing/model_coverage_metrics.py View File

@@ -172,8 +172,50 @@ class NeuronCoverage(CoverageMetrics):
float, the metric of 'neuron coverage'. float, the metric of 'neuron coverage'.


Examples: Examples:
>>> from mindspore.common.initializer import TruncatedNormal
>>> from mindspore.ops import operations as P
>>> from mindspore.train import Model
>>> from mindspore.ops import TensorSummary
>>> from mindarmour.fuzz_testing import NeuronCoverage
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self.conv1 = nn.Conv2d(1, 6, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.conv2 = nn.Conv2d(6, 16, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.fc1 = nn.Dense(16 * 5 * 5, 120, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc2 = nn.Dense(120, 84, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc3 = nn.Dense(84, 10, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.relu = nn.ReLU()
... self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
... self.reshape = P.Reshape()
... self.summary = TensorSummary()
... def construct(self, x):
... x = self.conv1(x)
... x = self.relu(x)
... self.summary('conv1', x)
... x = self.max_pool2d(x)
... x = self.conv2(x)
... x = self.relu(x)
... self.summary('conv2', x)
... x = self.max_pool2d(x)
... x = self.reshape(x, (-1, 16 * 5 * 5))
... x = self.fc1(x)
... x = self.relu(x)
... self.summary('fc1', x)
... x = self.fc2(x)
... x = self.relu(x)
... self.summary('fc2', x)
... x = self.fc3(x)
... self.summary('fc3', x)
... return x
>>> net = Net()
>>> model = Model(net)
>>> batch_size = 8
>>> num_classe = 10
>>> train_images = np.random.rand(32, 1, 32, 32).astype(np.float32)
>>> test_images = np.random.rand(batch_size, 1, 32, 32).astype(np.float32)
>>> nc = NeuronCoverage(model, threshold=0.1) >>> nc = NeuronCoverage(model, threshold=0.1)
>>> nc_metrics = nc.get_metrics(test_data)
>>> nc_metrics = nc.get_metrics(test_images)
""" """
dataset = check_numpy_param('dataset', dataset) dataset = check_numpy_param('dataset', dataset)
batches = math.ceil(dataset.shape[0] / self.batch_size) batches = math.ceil(dataset.shape[0] / self.batch_size)
@@ -219,8 +261,50 @@ class TopKNeuronCoverage(CoverageMetrics):
float, the metrics of 'top k neuron coverage'. float, the metrics of 'top k neuron coverage'.


Examples: Examples:
>>> from mindspore.common.initializer import TruncatedNormal
>>> from mindspore.ops import operations as P
>>> from mindspore.train import Model
>>> from mindspore.ops import TensorSummary
>>> from mindarmour.fuzz_testing import TopKNeuronCoverage
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self.conv1 = nn.Conv2d(1, 6, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.conv2 = nn.Conv2d(6, 16, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.fc1 = nn.Dense(16 * 5 * 5, 120, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc2 = nn.Dense(120, 84, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc3 = nn.Dense(84, 10, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.relu = nn.ReLU()
... self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
... self.reshape = P.Reshape()
... self.summary = TensorSummary()
... def construct(self, x):
... x = self.conv1(x)
... x = self.relu(x)
... self.summary('conv1', x)
... x = self.max_pool2d(x)
... x = self.conv2(x)
... x = self.relu(x)
... self.summary('conv2', x)
... x = self.max_pool2d(x)
... x = self.reshape(x, (-1, 16 * 5 * 5))
... x = self.fc1(x)
... x = self.relu(x)
... self.summary('fc1', x)
... x = self.fc2(x)
... x = self.relu(x)
... self.summary('fc2', x)
... x = self.fc3(x)
... self.summary('fc3', x)
... return x
>>> net = Net()
>>> model = Model(net)
>>> batch_size = 8
>>> num_classe = 10
>>> train_images = np.random.rand(32, 1, 32, 32).astype(np.float32)
>>> test_images = np.random.rand(batch_size, 1, 32, 32).astype(np.float32)
>>> tknc = TopKNeuronCoverage(model, top_k=3) >>> tknc = TopKNeuronCoverage(model, top_k=3)
>>> metrics = tknc.get_metrics(test_data)
>>> metrics = tknc.get_metrics(test_images)
""" """
dataset = check_numpy_param('dataset', dataset) dataset = check_numpy_param('dataset', dataset)
batches = math.ceil(dataset.shape[0] / self.batch_size) batches = math.ceil(dataset.shape[0] / self.batch_size)
@@ -269,8 +353,50 @@ class SuperNeuronActivateCoverage(CoverageMetrics):
float, the metric of 'strong neuron activation coverage'. float, the metric of 'strong neuron activation coverage'.


Examples: Examples:
>>> snac = SuperNeuronActivateCoverage(model, train_dataset)
>>> metrics = snac.get_metrics(test_data)
>>> from mindspore.common.initializer import TruncatedNormal
>>> from mindspore.ops import operations as P
>>> from mindspore.train import Model
>>> from mindspore.ops import TensorSummary
>>> from mindarmour.fuzz_testing import SuperNeuronActivateCoverage
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self.conv1 = nn.Conv2d(1, 6, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.conv2 = nn.Conv2d(6, 16, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.fc1 = nn.Dense(16 * 5 * 5, 120, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc2 = nn.Dense(120, 84, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc3 = nn.Dense(84, 10, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.relu = nn.ReLU()
... self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
... self.reshape = P.Reshape()
... self.summary = TensorSummary()
... def construct(self, x):
... x = self.conv1(x)
... x = self.relu(x)
... self.summary('conv1', x)
... x = self.max_pool2d(x)
... x = self.conv2(x)
... x = self.relu(x)
... self.summary('conv2', x)
... x = self.max_pool2d(x)
... x = self.reshape(x, (-1, 16 * 5 * 5))
... x = self.fc1(x)
... x = self.relu(x)
... self.summary('fc1', x)
... x = self.fc2(x)
... x = self.relu(x)
... self.summary('fc2', x)
... x = self.fc3(x)
... self.summary('fc3', x)
... return x
>>> net = Net()
>>> model = Model(net)
>>> batch_size = 8
>>> num_classe = 10
>>> train_images = np.random.rand(32, 1, 32, 32).astype(np.float32)
>>> test_images = np.random.rand(batch_size, 1, 32, 32).astype(np.float32)
>>> snac = SuperNeuronActivateCoverage(model, train_images)
>>> metrics = snac.get_metrics(test_images)
""" """
dataset = check_numpy_param('dataset', dataset) dataset = check_numpy_param('dataset', dataset)
if not self.incremental or not self._activate_table: if not self.incremental or not self._activate_table:
@@ -319,8 +445,50 @@ class NeuronBoundsCoverage(SuperNeuronActivateCoverage):
float, the metric of 'neuron boundary coverage'. float, the metric of 'neuron boundary coverage'.


Examples: Examples:
>>> nbc = NeuronBoundsCoverage(model, train_dataset)
>>> metrics = nbc.get_metrics(test_data)
>>> from mindspore.common.initializer import TruncatedNormal
>>> from mindspore.ops import operations as P
>>> from mindspore.train import Model
>>> from mindspore.ops import TensorSummary
>>> from mindarmour.fuzz_testing import NeuronBoundsCoverage
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self.conv1 = nn.Conv2d(1, 6, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.conv2 = nn.Conv2d(6, 16, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.fc1 = nn.Dense(16 * 5 * 5, 120, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc2 = nn.Dense(120, 84, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc3 = nn.Dense(84, 10, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.relu = nn.ReLU()
... self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
... self.reshape = P.Reshape()
... self.summary = TensorSummary()
... def construct(self, x):
... x = self.conv1(x)
... x = self.relu(x)
... self.summary('conv1', x)
... x = self.max_pool2d(x)
... x = self.conv2(x)
... x = self.relu(x)
... self.summary('conv2', x)
... x = self.max_pool2d(x)
... x = self.reshape(x, (-1, 16 * 5 * 5))
... x = self.fc1(x)
... x = self.relu(x)
... self.summary('fc1', x)
... x = self.fc2(x)
... x = self.relu(x)
... self.summary('fc2', x)
... x = self.fc3(x)
... self.summary('fc3', x)
... return x
>>> net = Net()
>>> model = Model(net)
>>> batch_size = 8
>>> num_classe = 10
>>> train_images = np.random.rand(32, 1, 32, 32).astype(np.float32)
>>> test_images = np.random.rand(batch_size, 1, 32, 32).astype(np.float32)
>>> nbc = NeuronBoundsCoverage(model, train_images)
>>> metrics = nbc.get_metrics(test_images)
""" """
dataset = check_numpy_param('dataset', dataset) dataset = check_numpy_param('dataset', dataset)
if not self.incremental or not self._activate_table: if not self.incremental or not self._activate_table:
@@ -383,8 +551,50 @@ class KMultisectionNeuronCoverage(SuperNeuronActivateCoverage):
float, the metric of 'k-multisection neuron coverage'. float, the metric of 'k-multisection neuron coverage'.


Examples: Examples:
>>> kmnc = KMultisectionNeuronCoverage(model, train_dataset, segmented_num=100)
>>> metrics = kmnc.get_metrics(test_data)
>>> from mindspore.common.initializer import TruncatedNormal
>>> from mindspore.ops import operations as P
>>> from mindspore.train import Model
>>> from mindspore.ops import TensorSummary
>>> from mindarmour.fuzz_testing import KMultisectionNeuronCoverage
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self.conv1 = nn.Conv2d(1, 6, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.conv2 = nn.Conv2d(6, 16, 5, padding=0, weight_init=TruncatedNormal(0.02), pad_mode="valid")
... self.fc1 = nn.Dense(16 * 5 * 5, 120, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc2 = nn.Dense(120, 84, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.fc3 = nn.Dense(84, 10, TruncatedNormal(0.02), TruncatedNormal(0.02))
... self.relu = nn.ReLU()
... self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
... self.reshape = P.Reshape()
... self.summary = TensorSummary()
... def construct(self, x):
... x = self.conv1(x)
... x = self.relu(x)
... self.summary('conv1', x)
... x = self.max_pool2d(x)
... x = self.conv2(x)
... x = self.relu(x)
... self.summary('conv2', x)
... x = self.max_pool2d(x)
... x = self.reshape(x, (-1, 16 * 5 * 5))
... x = self.fc1(x)
... x = self.relu(x)
... self.summary('fc1', x)
... x = self.fc2(x)
... x = self.relu(x)
... self.summary('fc2', x)
... x = self.fc3(x)
... self.summary('fc3', x)
... return x
>>> net = Net()
>>> model = Model(net)
>>> batch_size = 8
>>> num_classe = 10
>>> train_images = np.random.rand(32, 1, 32, 32).astype(np.float32)
>>> test_images = np.random.rand(batch_size, 1, 32, 32).astype(np.float32)
>>> kmnc = KMultisectionNeuronCoverage(model, train_images, segmented_num=100)
>>> metrics = kmnc.get_metrics(test_images)
""" """


dataset = check_numpy_param('dataset', dataset) dataset = check_numpy_param('dataset', dataset)


+ 0
- 9
mindarmour/privacy/diff_privacy/monitor/monitor.py View File

@@ -196,10 +196,6 @@ class RDPMonitor(Callback):


Returns: Returns:
int, the recommended maximum training epochs. int, the recommended maximum training epochs.

Examples:
>>> rdp = PrivacyMonitorFactory.create(policy='rdp', num_samples=100, batch_size=32)
>>> suggest_epoch = rdp.max_epoch_suggest()
""" """
if self._target_delta is not None and self._max_eps is None: if self._target_delta is not None and self._max_eps is None:
msg = 'max_eps should be consistent with target_delta, but got None.' msg = 'max_eps should be consistent with target_delta, but got None.'
@@ -450,11 +446,6 @@ class ZCDPMonitor(Callback):


Returns: Returns:
int, the recommended maximum training epochs. int, the recommended maximum training epochs.

Examples:
>>> zcdp = PrivacyMonitorFactory.create(policy='zcdp',
>>> num_samples=60000, batch_size=32)
>>> suggest_epoch = zcdp.max_epoch_suggest()
""" """
epoch = 1 epoch = 1
while epoch < 10000: while epoch < 10000:


+ 4
- 3
mindarmour/privacy/evaluation/attacker.py View File

@@ -120,7 +120,7 @@ def _attack_rf(features, labels, random_grid, n_jobs):
return rf_model return rf_model




def get_attack_model(features, labels, config, n_jobs=-1):
def _get_attack_model(features, labels, config, n_jobs=-1):
""" """
Get trained attack model specify by config. Get trained attack model specify by config.


@@ -139,10 +139,11 @@ def get_attack_model(features, labels, config, n_jobs=-1):
sklearn.BaseEstimator, trained model specify by config["method"]. sklearn.BaseEstimator, trained model specify by config["method"].


Examples: Examples:
>>> from mindarmour.privacy.evaluation.attacker import get_attack_model
>>> features = np.random.randn(10, 10) >>> features = np.random.randn(10, 10)
>>> labels = np.random.randint(0, 2, 10) >>> labels = np.random.randint(0, 2, 10)
>>> config = {"method": "knn", "params": {"n_neighbors": [3, 5, 7]}}
>>> attack_model = get_attack_model(features, labels, config)
>>> config = {"method": "knn", "params": {"n_neighbors": [3, 5]}}
>>> attack_model = _get_attack_model(features, labels, config)
""" """
features, labels = check_pair_numpy_param("features", features, "labels", labels) features, labels = check_pair_numpy_param("features", features, "labels", labels)
config = check_param_type("config", config, dict) config = check_param_type("config", config, dict)


+ 0
- 1
mindarmour/privacy/evaluation/inversion_attack.py View File

@@ -135,7 +135,6 @@ class ImageInversionAttack:
... loss_weights=[1, 0.2, 5]) ... loss_weights=[1, 0.2, 5])
>>> inversion_images = inversion_attack.generate(target_features, iters=10) >>> inversion_images = inversion_attack.generate(target_features, iters=10)
>>> evaluate_result = inversion_attack.evaluate(original_images, inversion_images) >>> evaluate_result = inversion_attack.evaluate(original_images, inversion_images)
>>> print(evaluate_result)
""" """
def __init__(self, network, input_shape, input_bound, loss_weights=(1, 0.2, 5)): def __init__(self, network, input_shape, input_bound, loss_weights=(1, 0.2, 5)):
self._network = check_param_type('network', network, Cell) self._network = check_param_type('network', network, Cell)


+ 4
- 5
mindarmour/privacy/evaluation/membership_inference.py View File

@@ -25,7 +25,7 @@ from mindspore import Tensor
from mindarmour.utils.logger import LogUtil from mindarmour.utils.logger import LogUtil
from mindarmour.utils._check_param import check_param_type, check_param_multi_types, \ from mindarmour.utils._check_param import check_param_type, check_param_multi_types, \
check_model, check_numpy_param check_model, check_numpy_param
from .attacker import get_attack_model
from .attacker import _get_attack_model
from ._check_config import verify_config_params from ._check_config import verify_config_params


LOGGER = LogUtil.get_instance() LOGGER = LogUtil.get_instance()
@@ -123,8 +123,7 @@ class MembershipInference:
... data = np.random.randn(batches * batch_size,1,10).astype(np.float32) ... data = np.random.randn(batches * batch_size,1,10).astype(np.float32)
... label = np.random.randint(0,10, batches * batch_size).astype(np.int32) ... label = np.random.randint(0,10, batches * batch_size).astype(np.int32)
... for i in range(batches): ... for i in range(batches):
... yield data[i*batch_size:(i+1)*batch_size],\
... label[i*batch_size:(i+1)*batch_size]
... yield data[i*batch_size:(i+1)*batch_size], label[i*batch_size:(i+1)*batch_size]
>>> class Net(Cell): >>> class Net(Cell):
... def __init__(self): ... def __init__(self):
... super(Net, self).__init__() ... super(Net, self).__init__()
@@ -143,7 +142,7 @@ class MembershipInference:
>>> config = [{ >>> config = [{
... "method": "KNN", ... "method": "KNN",
... "params": {"n_neighbors": [3, 5, 7],} ... "params": {"n_neighbors": [3, 5, 7],}
}]
... }]
>>> ds_train = ds.GeneratorDataset(dataset_generator, ["image", "label"]) >>> ds_train = ds.GeneratorDataset(dataset_generator, ["image", "label"])
>>> ds_test = ds.GeneratorDataset(dataset_generator, ["image", "label"]) >>> ds_test = ds.GeneratorDataset(dataset_generator, ["image", "label"])
>>> inference_model.train(ds_train, ds_test, config) >>> inference_model.train(ds_train, ds_test, config)
@@ -195,7 +194,7 @@ class MembershipInference:


features, labels = self._transform(dataset_train, dataset_test) features, labels = self._transform(dataset_train, dataset_test)
for config in attack_config: for config in attack_config:
self._attack_list.append(get_attack_model(features, labels, config, n_jobs=self._n_jobs))
self._attack_list.append(_get_attack_model(features, labels, config, n_jobs=self._n_jobs))




def eval(self, dataset_train, dataset_test, metrics): def eval(self, dataset_train, dataset_test, metrics):


+ 0
- 7
mindarmour/reliability/concept_drift/concept_drift_check_time_series.py View File

@@ -158,13 +158,6 @@ class ConceptDriftCheckTimeSeries:
- numpy.ndarray, the concept drift score of the example series. - numpy.ndarray, the concept drift score of the example series.
- float, the threshold to judge concept drift. - float, the threshold to judge concept drift.
- list, the location of the concept drift. - list, the location of the concept drift.

Examples：
>>> concept = ConceptDriftCheckTimeSeries(window_size=100, rolling_window=10,
... step=10, threshold_index=1.5, need_label=False)
>>> data_example = 5*np.random.rand(1000)
>>> data_example[200: 800] = 20*np.random.rand(600)
>>> score, drift_threshold, drift_location = concept.concept_check(data_example)
""" """
# data check # data check
data = _check_array_not_empty('data', data) data = _check_array_not_empty('data', data)


+ 1
- 2
mindarmour/reliability/model_fault_injection/fault_injection.py View File

@@ -59,8 +59,7 @@ class FaultInjector:
... data = np.random.randn(batches * batch_size,1,10).astype(np.float32) ... data = np.random.randn(batches * batch_size,1,10).astype(np.float32)
... label = np.random.randint(0,10, batches * batch_size).astype(np.int32) ... label = np.random.randint(0,10, batches * batch_size).astype(np.int32)
... for i in range(batches): ... for i in range(batches):
... yield data[i*batch_size:(i+1)*batch_size],\
... label[i*batch_size:(i+1)*batch_size]
... yield data[i*batch_size:(i+1)*batch_size], label[i*batch_size:(i+1)*batch_size]
>>> net = Net() >>> net = Net()
>>> model = Model(net) >>> model = Model(net)
>>> ds_eval = ds.GeneratorDataset(dataset_generator, ['image', 'label']) >>> ds_eval = ds.GeneratorDataset(dataset_generator, ['image', 'label'])


+ 49
- 10
mindarmour/utils/util.py View File

@@ -102,12 +102,24 @@ class WithLossCell(Cell):
loss_fn (Function): The loss function is used for computing loss. loss_fn (Function): The loss function is used for computing loss.


Examples: Examples:
>>> data = Tensor(np.ones([1, 1, 32, 32]).astype(np.float32)*0.01)
>>> label = Tensor(np.ones([1, 10]).astype(np.float32))
>>> net = NET()
>>> from mindspore import Tensor
>>> from mindarmour.utils.util import WithLossCell
>>> import mindspore.ops.operations as P
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self._softmax = P.Softmax()
... self._Dense = nn.Dense(10,10)
... self._squeeze = P.Squeeze(1)
... def construct(self, inputs):
... out = self._softmax(inputs)
... out = self._Dense(out)
... return self._squeeze(out)
>>> data = Tensor(np.ones([2, 1, 10]).astype(np.float32)*0.01)
>>> labels = Tensor(np.ones([2, 10]).astype(np.float32))
>>> net = Net()
>>> loss_fn = nn.SoftmaxCrossEntropyWithLogits() >>> loss_fn = nn.SoftmaxCrossEntropyWithLogits()
>>> loss_net = WithLossCell(net, loss_fn) >>> loss_net = WithLossCell(net, loss_fn)
>>> loss_out = loss_net(data, label)
""" """
def __init__(self, network, loss_fn): def __init__(self, network, loss_fn):
super(WithLossCell, self).__init__() super(WithLossCell, self).__init__()
@@ -138,9 +150,23 @@ class GradWrapWithLoss(Cell):
network (Cell): The target network to wrap. network (Cell): The target network to wrap.


Examples: Examples:
>>> data = Tensor(np.ones([1, 1, 32, 32]).astype(np.float32)*0.01)
>>> labels = Tensor(np.ones([1, 10]).astype(np.float32))
>>> net = NET()
>>> from mindspore import Tensor
>>> from mindarmour.utils import GradWrapWithLoss
>>> from mindarmour.utils.util import WithLossCell
>>> import mindspore.ops.operations as P
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self._softmax = P.Softmax()
... self._Dense = nn.Dense(10,10)
... self._squeeze = P.Squeeze(1)
... def construct(self, inputs):
... out = self._softmax(inputs)
... out = self._Dense(out)
... return self._squeeze(out)
>>> data = Tensor(np.ones([2, 1, 10]).astype(np.float32)*0.01)
>>> labels = Tensor(np.ones([2, 10]).astype(np.float32))
>>> net = Net()
>>> loss_fn = nn.SoftmaxCrossEntropyWithLogits() >>> loss_fn = nn.SoftmaxCrossEntropyWithLogits()
>>> loss_net = WithLossCell(net, loss_fn) >>> loss_net = WithLossCell(net, loss_fn)
>>> grad_all = GradWrapWithLoss(loss_net) >>> grad_all = GradWrapWithLoss(loss_net)
@@ -176,12 +202,25 @@ class GradWrap(Cell):
network (Cell): The target network to wrap. network (Cell): The target network to wrap.


Examples: Examples:
>>> data = Tensor(np.ones([1, 1, 32, 32]).astype(np.float32)*0.01)
>>> label = Tensor(np.ones([1, 10]).astype(np.float32))
>>> from mindspore import Tensor
>>> from mindarmour.utils import GradWrap
>>> from mindarmour.utils.util import WithLossCell
>>> import mindspore.ops.operations as P
>>> class Net(nn.Cell):
... def __init__(self):
... super(Net, self).__init__()
... self._softmax = P.Softmax()
... self._Dense = nn.Dense(10,10)
... self._squeeze = P.Squeeze(1)
... def construct(self, inputs):
... out = self._softmax(inputs)
... out = self._Dense(out)
... return self._squeeze(out)
>>> data = Tensor(np.ones([2, 1, 10]).astype(np.float32)*0.01)
>>> labels = Tensor(np.ones([2, 10]).astype(np.float32))
>>> num_classes = 10 >>> num_classes = 10
>>> sens = np.zeros((data.shape[0], num_classes)).astype(np.float32) >>> sens = np.zeros((data.shape[0], num_classes)).astype(np.float32)
>>> sens[:, 1] = 1.0 >>> sens[:, 1] = 1.0
>>> net = NET()
>>> wrap_net = GradWrap(net) >>> wrap_net = GradWrap(net)
>>> wrap_net(data, Tensor(sens)) >>> wrap_net(data, Tensor(sens))
""" """


+ 5
- 5
tests/ut/python/privacy/evaluation/test_attacker.py View File

@@ -18,7 +18,7 @@ import pytest


import numpy as np import numpy as np


from mindarmour.privacy.evaluation.attacker import get_attack_model
from mindarmour.privacy.evaluation.attacker import _get_attack_model




@pytest.mark.level0 @pytest.mark.level0
@@ -35,7 +35,7 @@ def test_get_knn_model():
"n_neighbors": [3, 5, 7], "n_neighbors": [3, 5, 7],
} }
} }
knn_attacker = get_attack_model(features, labels, config_knn, -1)
knn_attacker = _get_attack_model(features, labels, config_knn, -1)
pred = knn_attacker.predict(features) pred = knn_attacker.predict(features)
assert pred is not None assert pred is not None


@@ -54,7 +54,7 @@ def test_get_lr_model():
"C": np.logspace(-4, 2, 10), "C": np.logspace(-4, 2, 10),
} }
} }
lr_attacker = get_attack_model(features, labels, config_lr, -1)
lr_attacker = _get_attack_model(features, labels, config_lr, -1)
pred = lr_attacker.predict(features) pred = lr_attacker.predict(features)
assert pred is not None assert pred is not None


@@ -75,7 +75,7 @@ def test_get_mlp_model():
"alpha": [0.0001, 0.001, 0.01], "alpha": [0.0001, 0.001, 0.01],
} }
} }
mlpc_attacker = get_attack_model(features, labels, config_mlpc, -1)
mlpc_attacker = _get_attack_model(features, labels, config_mlpc, -1)
pred = mlpc_attacker.predict(features) pred = mlpc_attacker.predict(features)
assert pred is not None assert pred is not None


@@ -98,6 +98,6 @@ def test_get_rf_model():
"min_samples_leaf": [1, 2, 4], "min_samples_leaf": [1, 2, 4],
} }
} }
rf_attacker = get_attack_model(features, labels, config_rf, -1)
rf_attacker = _get_attack_model(features, labels, config_rf, -1)
pred = rf_attacker.predict(features) pred = rf_attacker.predict(features)
assert pred is not None assert pred is not None

Write Preview
Loading…
Cancel
Save