Increase version to 0.2.0 (#74)

* Remove tensorflow dependency if not using keras model
* Remove xgboost dependency if not using xgboost model
* Documentation updates

Signed-off-by: abigailt <abigailt@il.ibm.com>

apt/risk/data_assessment/attack_strategy_utils.py

@@ -9,21 +9,20 @@ from apt.utils.datasets import ArrayDataset
 
 class AttackStrategyUtils(abc.ABC):
     """
-        Abstract base class for common utilities of various privacy attack strategies.
+    Abstract base class for common utilities of various privacy attack strategies.
     """
     pass
 
 
 class KNNAttackStrategyUtils(AttackStrategyUtils):
     """
-         Common utilities for attack strategy based on KNN distances.
+    Common utilities for attack strategy based on KNN distances.
+
+    :param use_batches: Use batches with a progress meter or not when finding KNNs for query set.
+    :param batch_size: if use_batches=True, the size of batch_size should be > 0.
     """
 
     def __init__(self, use_batches: bool = False, batch_size: int = 10) -> None:
-        """
-        :param use_batches: Use batches with a progress meter or not when finding KNNs for query set
-        :param batch_size: if use_batches=True, the size of batch_size should be > 0
-        """
         self.use_batches = use_batches
         self.batch_size = batch_size
         if use_batches:
@@ -31,11 +30,18 @@ class KNNAttackStrategyUtils(AttackStrategyUtils):
                 raise ValueError(f"When using batching batch_size should be > 0, and not {batch_size}")
 
     def fit(self, knn_learner: NearestNeighbors, dataset: ArrayDataset):
+        """
+        Fit the KNN learner.
+
+        :param knn_learner: The KNN model to fit.
+        :param dataset: The training set to fit the model on.
+        """
         knn_learner.fit(dataset.get_samples())
 
     def find_knn(self, knn_learner: NearestNeighbors, query_samples: ArrayDataset, distance_processor=None):
         """
         Nearest neighbor search function.
+
         :param query_samples: query samples, to which nearest neighbors are to be found
         :param knn_learner: unsupervised learner for implementing neighbor searches, after it was fitted
         :param distance_processor: function for processing the distance into another more relevant metric per sample.

apt/risk/data_assessment/dataset_assessment_manager.py

@@ -15,6 +15,12 @@ from apt.utils.datasets import ArrayDataset
 
 @dataclass
 class DatasetAssessmentManagerConfig:
+    """
+    Configuration for DatasetAssessmentManager.
+
+    :param persist_reports: Whether to save assessment results to filesystem.
+    :param generate_plots: Whether to generate and visualize plots as part of assessment.
+    """
     persist_reports: bool = False
     generate_plots: bool = False
 
@@ -22,14 +28,13 @@ class DatasetAssessmentManagerConfig:
 class DatasetAssessmentManager:
     """
     The main class for running dataset assessment attacks.
+
+    :param config: Configuration parameters to guide the dataset assessment process
     """
     attack_scores_per_record_knn_probabilities: list[DatasetAttackScore] = []
     attack_scores_whole_dataset_knn_distance: list[DatasetAttackScore] = []
 
     def __init__(self, config: Optional[DatasetAssessmentManagerConfig] = DatasetAssessmentManagerConfig) -> None:
-        """
-        :param config: Configuration parameters to guide the dataset assessment process
-        """
         self.config = config
 
     def assess(self, original_data_members: ArrayDataset, original_data_non_members: ArrayDataset,
@@ -67,14 +72,17 @@ class DatasetAssessmentManager:
         return [score_gl, score_h]
 
     def dump_all_scores_to_files(self):
+        """
+        Save assessment results to filesystem.
+        """
         if self.config.persist_reports:
             results_log_file = "_results.log.csv"
-            self.dump_scores_to_file(self.attack_scores_per_record_knn_probabilities,
+            self._dump_scores_to_file(self.attack_scores_per_record_knn_probabilities,
                                      "per_record_knn_probabilities" + results_log_file, True)
-            self.dump_scores_to_file(self.attack_scores_whole_dataset_knn_distance,
+            self._dump_scores_to_file(self.attack_scores_whole_dataset_knn_distance,
                                      "whole_dataset_knn_distance" + results_log_file, True)
 
     @staticmethod
-    def dump_scores_to_file(attack_scores: list[DatasetAttackScore], filename: str, header: bool):
+    def _dump_scores_to_file(attack_scores: list[DatasetAttackScore], filename: str, header: bool):
         run_results_df = pd.DataFrame(attack_scores).drop('result', axis=1, errors='ignore')  # don't serialize result
         run_results_df.to_csv(filename, header=header, encoding='utf-8', index=False, mode='w')  # Overwrite

apt/risk/data_assessment/dataset_attack.py

@@ -16,32 +16,30 @@ from apt.utils.datasets import ArrayDataset
 
 class Config(abc.ABC):
     """
-        The base class for dataset attack configurations
+    The base class for dataset attack configurations
     """
     pass
 
 
 class DatasetAttack(abc.ABC):
     """
-         The interface for performing privacy attack for risk assessment of synthetic datasets to be used in AI model
-         training. The original data members (training data) and non-members (the holdout data) should be available.
-         For reliability, all the datasets should be preprocessed and normalized.
+     The interface for performing privacy attack for risk assessment of synthetic datasets to be used in AI model
+     training. The original data members (training data) and non-members (the holdout data) should be available.
+     For reliability, all the datasets should be preprocessed and normalized.
+
+     :param original_data_members: A container for the training original samples and labels,
+            only samples are used in the assessment
+     :param original_data_non_members: A container for the holdout original samples and labels,
+            only samples are used in the assessment
+     :param synthetic_data: A container for the synthetic samples and labels, only samples are used in the assessment
+     :param config: Configuration parameters to guide the assessment process
+     :param dataset_name: A name to identify the dataset under attack, optional
+     :param attack_strategy_utils: Utils for use with the attack strategy, optional
     """
 
     def __init__(self, original_data_members: ArrayDataset, original_data_non_members: ArrayDataset,
                  synthetic_data: ArrayDataset, config: Config, dataset_name: str,
                  attack_strategy_utils: Optional[AttackStrategyUtils] = None) -> None:
-        """
-        :param original_data_members: A container for the training original samples and labels,
-            only samples are used in the assessment
-        :param original_data_non_members: A container for the holdout original samples and labels,
-            only samples are used in the assessment
-        :param synthetic_data: A container for the synthetic samples and labels, only samples are used in the assessment
-        :param config: Configuration parameters to guide the assessment process
-        :param dataset_name: A name to identify the dataset under attack, optional
-        :param attack_strategy_utils: Utils for use with the attack strategy, optional
-        """
-
         self.original_data_members = original_data_members
         self.original_data_non_members = original_data_non_members
         self.synthetic_data = synthetic_data
@@ -52,7 +50,8 @@ class DatasetAttack(abc.ABC):
     @abc.abstractmethod
     def assess_privacy(self) -> DatasetAttackScore:
         """
-        Assess the privacy of the dataset
+        Assess the privacy of the dataset.
+
         :return:
             score: DatasetAttackScore the privacy attack risk score
         """
@@ -61,14 +60,15 @@ class DatasetAttack(abc.ABC):
 
 class DatasetAttackMembership(DatasetAttack):
     """
-         An abstract base class for performing privacy risk assessment for synthetic datasets on a per-record level.
+    An abstract base class for performing privacy risk assessment for synthetic datasets on a per-record level.
     """
 
     @abc.abstractmethod
     def calculate_privacy_score(self, dataset_attack_result: DatasetAttackResultMembership,
                                 generate_plot: bool = False) -> DatasetAttackScore:
         """
-        Calculate dataset privacy score based on the result of the privacy attack
+        Calculate dataset privacy score based on the result of the privacy attack.
+
         :return:
             score: DatasetAttackScore
         """
@@ -78,15 +78,16 @@ class DatasetAttackMembership(DatasetAttack):
     def plot_roc_curve(dataset_name: str, member_probabilities: np.ndarray, non_member_probabilities: np.ndarray,
                        filename_prefix: str = ""):
         """
-        Plot ROC curve
-        :param dataset_name: dataset name, will become part of the plot filename
-        :param member_probabilities: probability estimates of the member samples, the training data
-        :param non_member_probabilities: probability estimates of the non-member samples, the hold-out data
-        :param filename_prefix: name prefix for the ROC curve plot
+        Plot ROC curve.
+
+        :param dataset_name: dataset name, will become part of the plot filename.
+        :param member_probabilities: probability estimates of the member samples, the training data.
+        :param non_member_probabilities: probability estimates of the non-member samples, the hold-out data.
+        :param filename_prefix: name prefix for the ROC curve plot.
         """
         labels = np.concatenate((np.zeros((len(non_member_probabilities),)), np.ones((len(member_probabilities),))))
         results = np.concatenate((non_member_probabilities, member_probabilities))
-        svc_disp = RocCurveDisplay.from_predictions(labels, results)
+        RocCurveDisplay.from_predictions(labels, results)
         plt.plot([0, 1], [0, 1], color="navy", linewidth=2, linestyle="--", label='No skills')
         plt.title('ROC curve')
         plt.savefig(f'{filename_prefix}{dataset_name}_roc_curve.png')
@@ -94,9 +95,10 @@ class DatasetAttackMembership(DatasetAttack):
     @staticmethod
     def calculate_metrics(member_probabilities: np.ndarray, non_member_probabilities: np.ndarray):
         """
-        Calculate attack performance metrics
-        :param member_probabilities: probability estimates of the member samples, the training data
-        :param non_member_probabilities: probability estimates of the non-member samples, the hold-out data
+        Calculate attack performance metrics.
+
+        :param member_probabilities: probability estimates of the member samples, the training data.
+        :param non_member_probabilities: probability estimates of the non-member samples, the hold-out data.
         :return:
             fpr: False Positive rate
             tpr: True Positive rate

apt/risk/data_assessment/dataset_attack_membership_knn_probabilities.py

@@ -19,18 +19,18 @@ from apt.utils.datasets import ArrayDataset
 
 @dataclass
 class DatasetAttackConfigMembershipKnnProbabilities(Config):
-    """Configuration for DatasetAttackMembershipKnnProbabilities.
+    """
+    Configuration for DatasetAttackMembershipKnnProbabilities.
 
-    Attributes:
-        k: Number of nearest neighbors to search
-        use_batches: Divide query samples into batches or not.
-        batch_size:  Query sample batch size.
-        compute_distance: A callable function, which takes two arrays representing 1D vectors as inputs and must return
-            one value indicating the distance between those vectors.
-            See 'metric' parameter in sklearn.neighbors.NearestNeighbors documentation.
-        distance_params:  Additional keyword arguments for the distance computation function, see 'metric_params' in
-            sklearn.neighbors.NearestNeighbors documentation.
-        generate_plot: Generate or not an AUR ROC curve and persist it in a file
+    :param k: Number of nearest neighbors to search.
+    :param use_batches: Divide query samples into batches or not.
+    :param batch_size:  Query sample batch size.
+    :param compute_distance: A callable function, which takes two arrays representing 1D vectors as inputs and must
+                             return one value indicating the distance between those vectors.
+                             See 'metric' parameter in sklearn.neighbors.NearestNeighbors documentation.
+    :param distance_params:  Additional keyword arguments for the distance computation function, see 'metric_params' in
+                             sklearn.neighbors.NearestNeighbors documentation.
+    :param generate_plot: Generate or not an AUR ROC curve and persist it in a file.
     """
     k: int = 5
     use_batches: bool = False
@@ -42,7 +42,14 @@ class DatasetAttackConfigMembershipKnnProbabilities(Config):
 
 @dataclass
 class DatasetAttackScoreMembershipKnnProbabilities(DatasetAttackScore):
-    """DatasetAttackMembershipKnnProbabilities privacy risk score.
+    """
+    DatasetAttackMembershipKnnProbabilities privacy risk score.
+
+    :param dataset_name: dataset name to be used in reports
+    :param roc_auc_score: the area under the receiver operating characteristic curve (AUC ROC) to evaluate the
+                          attack performance.
+    :param average_precision_score: the proportion of predicted members that are correctly members.
+    :param result: the result of the membership inference attack.
     """
     roc_auc_score: float
     average_precision_score: float
@@ -50,13 +57,6 @@ class DatasetAttackScoreMembershipKnnProbabilities(DatasetAttackScore):
 
     def __init__(self, dataset_name: str, roc_auc_score: float, average_precision_score: float,
                  result: DatasetAttackResultMembership) -> None:
-        """
-        dataset_name:    dataset name to be used in reports
-        roc_auc_score:   the area under the receiver operating characteristic curve (AUC ROC) to evaluate the attack
-                          performance.
-        average_precision_score: the proportion of predicted members that are correctly members
-        result:          the result of the membership inference attack
-        """
         super().__init__(dataset_name=dataset_name, risk_score=roc_auc_score, result=result)
         self.roc_auc_score = roc_auc_score
         self.average_precision_score = average_precision_score
@@ -64,24 +64,23 @@ class DatasetAttackScoreMembershipKnnProbabilities(DatasetAttackScore):
 
 class DatasetAttackMembershipKnnProbabilities(DatasetAttackMembership):
     """
-         Privacy risk assessment for synthetic datasets based on Black-Box MIA attack using distances of
-         members (training set) and non-members (holdout set) from their nearest neighbors in the synthetic dataset.
-         By default, the Euclidean distance is used (L2 norm), but another ``compute_distance()`` method can be provided
-         in configuration instead.
-         The area under the receiver operating characteristic curve (AUC ROC) gives the privacy risk measure.
+    Privacy risk assessment for synthetic datasets based on Black-Box MIA attack using distances of
+    members (training set) and non-members (holdout set) from their nearest neighbors in the synthetic dataset.
+    By default, the Euclidean distance is used (L2 norm), but another ``compute_distance()`` method can be provided
+    in configuration instead.
+    The area under the receiver operating characteristic curve (AUC ROC) gives the privacy risk measure.
+
+    :param original_data_members: A container for the training original samples and labels
+    :param original_data_non_members: A container for the holdout original samples and labels
+    :param synthetic_data: A container for the synthetic samples and labels
+    :param config: Configuration parameters to guide the attack, optional
+    :param dataset_name: A name to identify this dataset, optional
     """
 
     def __init__(self, original_data_members: ArrayDataset, original_data_non_members: ArrayDataset,
                  synthetic_data: ArrayDataset,
                  config: DatasetAttackConfigMembershipKnnProbabilities = DatasetAttackConfigMembershipKnnProbabilities(),
                  dataset_name: str = DEFAULT_DATASET_NAME):
-        """
-        :param original_data_members: A container for the training original samples and labels
-        :param original_data_non_members: A container for the holdout original samples and labels
-        :param synthetic_data: A container for the synthetic samples and labels
-        :param config: Configuration parameters to guide the attack, optional
-        :param dataset_name: A name to identify this dataset, optional
-        """
         attack_strategy_utils = KNNAttackStrategyUtils(config.use_batches, config.batch_size)
         super().__init__(original_data_members, original_data_non_members, synthetic_data, config, dataset_name,
                          attack_strategy_utils)
@@ -103,10 +102,9 @@ class DatasetAttackMembershipKnnProbabilities(DatasetAttackMembership):
         by the Parzen window density estimation in ``probability_per_sample()``, computed from the NN distances from the
         query samples to the synthetic data samples.
 
-        :return:
-            Privacy score of the attack together with the attack result with the probabilities of member and
-            non-member samples to be generated by the synthetic data generator based on the NN distances from the
-            query samples to the synthetic data samples
+        :return: Privacy score of the attack together with the attack result with the probabilities of member and
+                 non-member samples to be generated by the synthetic data generator based on the NN distances from the
+                 query samples to the synthetic data samples
         """
         # nearest neighbor search
         self.attack_strategy_utils.fit(self.knn_learner, self.synthetic_data)
@@ -130,11 +128,11 @@ class DatasetAttackMembershipKnnProbabilities(DatasetAttackMembership):
         """
         Evaluate privacy score from the probabilities of member and non-member samples to be generated by the synthetic
         data generator. The probabilities are computed by the ``assess_privacy()`` method.
-        :param dataset_attack_result attack result containing probabilities of member and non-member samples to be
-                generated by the synthetic data generator
-        :param generate_plot generate AUC ROC curve plot and persist it
-        :return:
-            score of the attack, based on distance-based probabilities - mainly the ROC AUC score
+
+        :param dataset_attack_result: attack result containing probabilities of member and non-member samples to be
+                generated by the synthetic data generator.
+        :param generate_plot: generate AUC ROC curve plot and persist it.
+        :return: score of the attack, based on distance-based probabilities - mainly the ROC AUC score.
         """
         member_proba, non_member_proba = \
             dataset_attack_result.member_probabilities, dataset_attack_result.non_member_probabilities
@@ -151,10 +149,10 @@ class DatasetAttackMembershipKnnProbabilities(DatasetAttackMembership):
         """
         For every sample represented by its distance from the query sample to its KNN in synthetic data,
         computes the probability of the synthetic data to be part of the query dataset.
+
         :param distances: distance between every query sample in batch to its KNNs among synthetic samples, a numpy
-        array of size (n, k) with n being the number of samples, k - the number of KNNs
-        :return:
-            probability estimates of the query samples being generated and so - of being part of the synthetic set, a
-            numpy array of size (n,)
+                          array of size (n, k) with n being the number of samples, k - the number of KNNs.
+        :return: probability estimates of the query samples being generated and so - of being part of the synthetic set,
+                 a numpy array of size (n,)
         """
         return np.average(np.exp(-distances), axis=1)

apt/risk/data_assessment/dataset_attack_result.py

@@ -8,11 +8,21 @@ DEFAULT_DATASET_NAME = "dataset"
 
 @dataclass
 class DatasetAttackResult:
+    """
+    Basic class for storing privacy risk assessment results.
+    """
     pass
 
 
 @dataclass
 class DatasetAttackScore:
+    """
+    Basic class for storing privacy risk assessment scores.
+
+    :param dataset_name: The name of the dataset that was assessed.
+    :param risk_score: The privacy risk score.
+    :param result: An optional list of more detailed results.
+    """
     dataset_name: str
     risk_score: float
     result: Optional[DatasetAttackResult]
@@ -20,5 +30,11 @@ class DatasetAttackScore:
 
 @dataclass
 class DatasetAttackResultMembership(DatasetAttackResult):
+    """
+    Class for storing membership attack results.
+
+    :param member_probabilities: The attack probabilities for member samples.
+    :param non_member_probabilities: The attack probabilities for non-member samples.
+    """
     member_probabilities: np.ndarray
     non_member_probabilities: np.ndarray

apt/risk/data_assessment/dataset_attack_whole_dataset_knn_distance.py

@@ -20,16 +20,16 @@ K = 1  # Number of nearest neighbors to search. For DCR we need only the nearest
 
 @dataclass
 class DatasetAttackConfigWholeDatasetKnnDistance(Config):
-    """Configuration for DatasetAttackWholeDatasetKnnDistance.
+    """
+    Configuration for DatasetAttackWholeDatasetKnnDistance.
 
-    Attributes:
-        use_batches:  Divide query samples into batches or not.
-        batch_size:   Query sample batch size.
-        compute_distance: A callable function, which takes two arrays representing 1D vectors as inputs and must return
-            one value indicating the distance between those vectors.
-            See 'metric' parameter in sklearn.neighbors.NearestNeighbors documentation.
-        distance_params:  Additional keyword arguments for the distance computation function, see 'metric_params' in
-            sklearn.neighbors.NearestNeighbors documentation.
+    :param use_batches: Divide query samples into batches or not.
+    :param batch_size: Query sample batch size.
+    :param compute_distance: A callable function, which takes two arrays representing 1D vectors as inputs and must
+                             return one value indicating the distance between those vectors.
+                             See 'metric' parameter in sklearn.neighbors.NearestNeighbors documentation.
+    :param distance_params:  Additional keyword arguments for the distance computation function, see 'metric_params' in
+                             sklearn.neighbors.NearestNeighbors documentation.
     """
     use_batches: bool = False
     batch_size: int = 10
@@ -39,41 +39,40 @@ class DatasetAttackConfigWholeDatasetKnnDistance(Config):
 
 @dataclass
 class DatasetAttackScoreWholeDatasetKnnDistance(DatasetAttackScore):
-    """DatasetAttackWholeDatasetKnnDistance privacy risk score.
+    """
+    DatasetAttackWholeDatasetKnnDistance privacy risk score.
+
+    :param dataset_name: Dataset name to be used in reports.
+    :param share: The share of synthetic records closer to the training than the holdout dataset.
+                  A value of 0.5 or close to it means good privacy.
     """
     share: float
     assessment_type: str = 'WholeDatasetKnnDistance'  # to be used in reports
 
     def __init__(self, dataset_name: str, share: float) -> None:
-        """
-        dataset_name:    dataset name to be used in reports
-        share : the share of synthetic records closer to the training than the holdout dataset.
-                A value of 0.5 or close to it means good privacy.
-        """
         super().__init__(dataset_name=dataset_name, risk_score=share, result=None)
         self.share = share
 
 
 class DatasetAttackWholeDatasetKnnDistance(DatasetAttack):
     """
-         Privacy risk assessment for synthetic datasets based on distances of synthetic data records from
-         members (training set) and non-members (holdout set). The privacy risk measure is the share of synthetic
-         records closer to the training than the holdout dataset.
-         By default, the Euclidean distance is used (L2 norm), but another compute_distance() method can be provided in
-         configuration instead.
+    Privacy risk assessment for synthetic datasets based on distances of synthetic data records from
+    members (training set) and non-members (holdout set). The privacy risk measure is the share of synthetic
+    records closer to the training than the holdout dataset.
+    By default, the Euclidean distance is used (L2 norm), but another compute_distance() method can be provided in
+    configuration instead.
+
+    :param original_data_members: A container for the training original samples and labels.
+    :param original_data_non_members: A container for the holdout original samples and labels.
+    :param synthetic_data: A container for the synthetic samples and labels.
+    :param config: Configuration parameters to guide the assessment process, optional.
+    :param dataset_name: A name to identify this dataset, optional.
     """
 
     def __init__(self, original_data_members: ArrayDataset, original_data_non_members: ArrayDataset,
                  synthetic_data: ArrayDataset,
                  config: DatasetAttackConfigWholeDatasetKnnDistance = DatasetAttackConfigWholeDatasetKnnDistance(),
                  dataset_name: str = DEFAULT_DATASET_NAME):
-        """
-        :param original_data_members: A container for the training original samples and labels
-        :param original_data_non_members: A container for the holdout original samples and labels
-        :param synthetic_data: A container for the synthetic samples and labels
-        :param config: Configuration parameters to guide the assessment process, optional
-        :param dataset_name: A name to identify this dataset, optional
-        """
         attack_strategy_utils = KNNAttackStrategyUtils(config.use_batches, config.batch_size)
         super().__init__(original_data_members, original_data_non_members, synthetic_data, config, dataset_name,
                          attack_strategy_utils)
@@ -90,6 +89,7 @@ class DatasetAttackWholeDatasetKnnDistance(DatasetAttack):
         """
         Calculate the share of synthetic records closer to the training than the holdout dataset, based on the
         DCR computed by 'calculate_distances()'.
+
         :return:
             score of the attack, based on the NN distances from the query samples to the synthetic data samples
         """