API Reference¶

This is the API documentation for PySAD.

Core¶

The pysad.core module covers base classes of the PySAD.

`core.BaseMetric`	Abstract base class for metrics.
`core.BaseModel`	Abstract base class for the models.
`core.BasePostprocessor`	Base class for postprocessing methods.
`core.BaseStatistic`	Abstact base class for the statistics.
`core.BaseStreamer`	Abstract base class to simulate the streaming data.
`core.BaseTransformer`	Base class for transforming methods.

Individual Anomaly Models¶

The pysad.models module includes anomaly detection models to score the anomalousness of instances.

`models.ExactStorm`	The Exact-STORM method [BAF07].
`models.HalfSpaceTrees`	Half-Space Trees method [BTTL11].
`models.IForestASD`	An Anomaly Detection Approach Based on Isolation Forest Algorithm for Streaming Data using Sliding Window [BDF13].
`models.KitNet`	KitNET is a lightweight online anomaly detection algorithm based on an ensemble of autoencoders [BMDES18].
`models.KNNCAD`	Conformalized density- and distance-based anomaly detection in time-series data [BBI16], which uses a combination of a feature extraction method, an approach to assess a score whether a new observation differs significantly from a previously observed data, and a probabilistic interpretation of this score based on the conformal paradigm.
`models.LODA`	The LODA model [BPevny16] The implemnetation is adapted to the steraming framework from the PyOD framework.
`models.LocalOutlierProbability`	The implementation of streaming Local Outlier Probabilities method [BKKrogerSZ09], which uses the implementation of PyNomaly library [BCon18].
`models.MedianAbsoluteDeviation`	Median Absolute Deviation method [BHVK17].
`models.NullModel`	The model that returns 0.5 for all instances, which is added for testing and pipelining convenience purposes.
`models.PerfectModel`	This model directly outputs the ground truth labels.
`models.RandomModel`	Random scorer that chooses a score between 0 and 1 ignoring the input.
`models.RelativeEntropy`	Relative entropy based anomaly detection model on univariate stream [BALPA17].
`models.RobustRandomCutForest`	Robust Random Cut Forest model [BGMRS16].
`models.RSHash`	Subspace outlier detection in linear time with randomized hashing [BSA16].
`models.StandardAbsoluteDeviation`	The model that assigns the deviation from the mean (or median) and divides with the standard deviation.
`models.xStream`	The xStream model for row-streaming data [BMLA18].

Integration Models¶

The pysad.models.integrations module contains models to integrate batch anomaly detection models to the streaming setting.

`models.integrations.ReferenceWindowModel`	This PyOD model wrapper wraps the batch anomaly detectors.
`models.integrations.OneFitModel`	The wrapper model fits the model_cls to the initial instnaces.

Score Ensemblers¶

The pysad.transform.ensemble module consist of ensemblers to combine scores from multiple anomaly detectors.

`transform.ensemble.MaximumScoreEnsembler`	An ensembler that results the maximum of the previous scores.
`transform.ensemble.MedianScoreEnsembler`	An ensembler that results the median of the previous scores.
`transform.ensemble.AverageScoreEnsembler`	An wrapper class that results in the weighted average of the anomaly scores from multiple anomaly detectors.
`transform.ensemble.MaximumOfAverageScoreEnsembler`	Maximum of average scores ensembler that outputs the maximum of average.
`transform.ensemble.AverageOfMaximumScoreEnsembler`	Maximum of average scores ensembler that outputs the maximum of average.

Probability Calibrators¶

The pysad.transform.probability_calibration module includes probability calibrators to convert module scores into true probabilities for decision-making on anomalousness.

`transform.probability_calibration.ConformalProbabilityCalibrator`	This class provides an interface to convert the scores into probabilities through conformal prediction.
`transform.probability_calibration.GaussianTailProbabilityCalibrator`	Assuming that the scores follow normal distribution, this class provides an interface to convert the scores into probabilities via Q-function, i.e., the tail function of Gaussian distribution [BALPA17].

Projectors¶

The pysad.transform.projection module contains methods to project input into (possibly) lower dimensional space to better discriminate anomalies.

`transform.projection.StreamhashProjector`	Streamhash projection method from Manzoor et.
`transform.projection.GaussianRandomProjector`	Reduces dimensionality through Gaussian random projection.
`transform.projection.SparseRandomProjector`	The wrapper method for Sklearn's SparseRandomProjection.

Preprocessors¶

” The pysad.transform.preprocessing module includes preprocessing methods to transform inputs such as normalizers.

`transform.preprocessing.IdentityScaler`	A scaler that does not modify the input, which is added for convenience.
`transform.preprocessing.InstanceStandardScaler`	Standard deviation scaling per instance.
`transform.preprocessing.InstanceUnitNormScaler`	A scaler that makes the instance feature vector's norm equal to 1, i.e., the unit vector.

Postprocessors¶

The pysad.transform.postprocessing module includes postprocessors to transform model scores for streaming learning.

`pysad.transform.postprocessing.AveragePostprocessor`	A postprocessor that convert a score to the average of of all previous scores.
`pysad.transform.postprocessing.MaxPostprocessor`	A postprocessor that convert a score to the maximum of of all previous scores.
`pysad.transform.postprocessing.MedianPostprocessor`	A postprocessor that convert a score to the median of of all previous scores.
`pysad.transform.postprocessing.ZScorePostprocessor`	A postprocessor that normalize the score via Z-score normalization.
`pysad.transform.postprocessing.RunningAveragePostprocessor`	A postprocessor that convert a score to the average of of all previous scores in the window.
`pysad.transform.postprocessing.RunningMaxPostprocessor`	A postprocessor that convert a score to the maximum of of all previous scores in the window.
`pysad.transform.postprocessing.RunningMedianPostprocessor`	A postprocessor that convert a score to the median of of all previous scores in the window.
`pysad.transform.postprocessing.RunningZScorePostprocessor`	A postprocessor that normalizes score using Z-score normalization with the statistics of the window.

Statistics¶

The pysad.statistics module contains methods to keep track of statistics on streaming data.

`statistics.AbsStatistic`	The absolute value of the statistic that is tracked.
`statistics.RunningStatistic`	The running statistic that wraps any other statistics to track statistics with a fixed window size.
`statistics.AverageMeter`	The average of the values.
`statistics.CountMeter`	A simple counter statistic.
`statistics.MaxMeter`	The statistic that keeps track of the maximum value.
`statistics.MedianMeter`	The statistic that keeps track of the median.
`statistics.MinMeter`	The statistic that keeps track of the minimum value.
`statistics.SumMeter`	The statistic that keeps track of the sum of values.
`statistics.SumSquaresMeter`	The statistic that keeps track of the sum of squares.
`statistics.VarianceMeter`	The statistic that keeps track of the variance of the values.

Evaluators¶

The pysad.evaluation module includes evaluation metrics for anomaly detection on streaming data.

`evaluation.BaseSKLearnMetric`	Abstract base class to wrap the sklearn metrics.
`evaluation.PrecisionMetric`	Precision wrapper class for sklearn.
`evaluation.RecallMetric`	Recall wrapper class for sklearn.
`evaluation.AUROCMetric`	Area under roc curve wrapper class for sklearn.
`evaluation.AUPRMetric`	Area under PR curve wrapper class for sklearn.
`evaluation.WindowedMetric`	A helper class to evaluate windowed metrics.

Utilities¶

The pysad.utils module includes utility functions used in the PySAD framework, which can also be useful streaming learning.

`utils.Window`	Window to limit the instances in list and keep the size fixed when full.
`utils.Data`	A helper class to load various data.
`utils.ArrayStreamer`	Simulator class to iterate array(s).
`utils.PandasStreamer`	Simulator class to iterate dataframe(s).
`utils._iterate`	Iterates array of features and possibly labels.
`utils.get_minmax_array`	Utility method that returns the boundaries for each feature of the input array.
`utils.get_minmax_scalar`	Utility method that returns the boundaries of the input array.
`utils.fix_seed`	Utility method to fix the seed for randomness.