API Reference

This is an overview of the API.

skactiveml.base:

The skactiveml.base package implements the base classes for skactiveml.

Classes

base.QueryStrategy	Base class for all query strategies in scikit-activeml.
base.SingleAnnotatorPoolQueryStrategy	Base class for all pool-based active learning query strategies with a single annotator in scikit-activeml.
base.MultiAnnotatorPoolQueryStrategy	Base class for all pool-based active learning query strategies with multiple annotators in scikit-activeml.
base.BudgetManager	Base class for all budget managers for stream-based active learning in scikit-activeml to model budgeting constraints.
base.SingleAnnotatorStreamQueryStrategy	Base class for all stream-based active learning query strategies in scikit-activeml.
base.SkactivemlClassifier	Base class for scikit-activeml classifiers such that missing labels, user-defined classes, and cost-sensitive classification (i.e., cost matrix) can be handled.
base.ClassFrequencyEstimator	Extends scikit-activeml classifiers to estimators that are able to estimate class frequencies for given samples (by calling 'predict_freq').
base.AnnotatorModelMixin	Base class of all annotator models estimating the performances of annotators for given samples.
base.SkactivemlRegressor	Base class for scikit-activeml regressors.
base.ProbabilisticRegressor	Base class for scikit-activeml probabilistic regressors.

skactiveml.pool:

The skactiveml.pool package implements query strategies for pool-based active learning.

Classes

pool.RandomSampling	Random Sampling.
pool.ProbabilisticAL	(Multi-class) Probabilistic Active Learning
pool.UncertaintySampling	Uncertainty Sampling.
pool.EpistemicUncertaintySampling	Epistemic Uncertainty Sampling.
pool.MonteCarloEER	This class implements the expected error method from [1] that uses a Monte-Carlo approach to estimate the error.
pool.ValueOfInformationEER	This class implements the expected error method from [1] that estimates the value of information.
pool.QueryByCommittee	Query-by-Committee (QBC)
pool.Quire	Implementation of the AL strategy "QUerying Informative and Representative Examples" (QUIRE for short) [1].
pool.FourDs	Implementation of the pool-based query strategy 4DS for training a MixtureModelClassifier [1].
pool.CostEmbeddingAL	Active Learning with Cost Embedding (ALCE).
pool.ExpectedModelChangeMaximization	Expected Model Change.
pool.ExpectedModelOutputChange	Regression based Expected Model Output Change.
pool.ExpectedModelVarianceReduction	Expected Model Variance Reduction.
pool.KLDivergenceMaximization	Regression based Kullback Leibler Divergence Maximization.
pool.GreedySamplingX	Greedy Sampling on the feature space.
pool.GreedySamplingTarget	Greedy Sampling on the target space.
pool.DiscriminativeAL	Discriminative Active Learning.
pool.BatchBALD	Batch Bayesian Active Learning by Disagreement (BatchBALD)
pool.Clue	Clustering Uncertainty-weighted Embeddings (CLUE)
pool.CoreSet	Core Set
pool.TypiClust	Typical Clustering
pool.Badge	Batch Active Learning by Diverse Gradient Embedding (BADGE)
pool.ProbCover	Probability Coverage
pool.ContrastiveAL	Contrastive Active Learning (ContrastiveAL)
pool.GreedyBALD	Greedy Bayesian Active Learning by Disagreement (GreedyBALD)
pool.RegressionTreeBasedAL	Regression Tree-based Active Learning
pool.SubSamplingWrapper	Sub-sampling Wrapper.
pool.ParallelUtilityEstimationWrapper	Parallel Utility Estimation Wrapper.

Functions

pool.cost_reduction	Calculate the expected cost reduction.
pool.uncertainty_scores	Computes uncertainty scores.
pool.expected_average_precision	Calculate the expected average precision.
pool.average_kl_divergence	Calculates the average Kullback-Leibler (KL) divergence for measuring the level of disagreement in QueryByCommittee.
pool.vote_entropy	Calculates the vote entropy for measuring the level of disagreement in QueryByCommittee.
pool.variation_ratios	Calculates the variation ratios for measuring the level of disagreement in QueryByCommittee.
pool.batch_bald	BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning
pool.k_greedy_center	An active learning method that greedily forms a batch to minimize the maximum distance to a cluster center among all unlabeled datapoints.

skactiveml.pool.multiannotator:

The skactiveml.pool.multiannotator package implements multi annotator pool-based active learning for multiple annotators.

Classes

pool.multiannotator.IntervalEstimationThreshold	The strategy 'Interval Estimation Threshold' (IEThresh) [1] is useful for addressing the exploration vs.
pool.multiannotator.IntervalEstimationAnnotModel	This annotator model relies on 'Interval Estimation Learning' (IELearning) for estimating the annotation performances, i.e., labeling accuracies, of multiple annotators [1].
pool.multiannotator.SingleAnnotatorWrapper	Implementation of a wrapper class for pool-based active learning query strategies with a single annotator such that it transforms the query strategy for the single annotator into a query strategy for multiple annotators by choosing an annotator randomly or according to the parameter A_pef and setting the labeled matrix to a labeled vector by an aggregation function, e.g., majority voting.

skactiveml.pool.utils:

Classes

pool.utils.IndexClassifierWrapper

Classifier to simplify retraining classifiers in an active learning scenario.

skactiveml.stream:

The skactiveml.stream module implements query strategies for stream-based active learning.

Classes

stream.StreamRandomSampling	Random Sampling for Datastreams.
stream.PeriodicSampling	The PeriodicSampling samples instances periodically.
stream.FixedUncertainty	The FixedUncertainty (Fixed-Uncertainty in [1]) query strategy samples instances based on the classifiers uncertainty assessed based on the classifier's predictions.
stream.VariableUncertainty	The VariableUncertainty (Var-Uncertainty in [1]) query strategy samples instances based on the classifiers uncertainty assessed based on the classifier's predictions.
stream.Split	The Split [1] query strategy samples in 100*v% of instances randomly and in 100*(1-v)% of cases according to VariableUncertainty.
stream.StreamProbabilisticAL	Probabilistic Active Learning in Datastreams (StreamProbabilisticAL) is an extension to Multi-Class Probabilistic Active Learning (McPAL) (see pool.ProbabilisticAL).
stream.RandomVariableUncertainty	The RandomVariableUncertainty (Ran-Var-Uncertainty in [1]) query strategy samples instances based on the classifier's uncertainty assessed based on the classifier's predictions.
stream.StreamDensityBasedAL	The StreamDensityBasedAL [1] query strategy is an extension to the uncertainty based query strategies proposed by Žliobaitė et al. [2].
stream.CognitiveDualQueryStrategy	This class is the base for the CognitiveDualQueryStrategy query strategy proposed in [1].
stream.CognitiveDualQueryStrategyRan	This class implements the CognitiveDualQueryStrategy strategy with Random Sampling.
stream.CognitiveDualQueryStrategyRanVarUn	This class implements the CognitiveDualQueryStrategy strategy with RandomVariableUncertainty.
stream.CognitiveDualQueryStrategyVarUn	This class implements the CognitiveDualQueryStrategy strategy with VariableUncertainty.
stream.CognitiveDualQueryStrategyFixUn	This class implements the CognitiveDualQueryStrategy strategy with FixedUncertainty.

skactiveml.stream.budgetmanager:

The skactiveml.stream.budgetmanager module implements budget managers, which are capable of modeling the budget constraints in stream-based active learning settings.

Classes

stream.budgetmanager.EstimatedBudgetZliobaite	Budget manager which checks, whether the specified budget has been exhausted already.
stream.budgetmanager.FixedUncertaintyBudgetManager	Budget manager which is optimized for FixedUncertainty and checks, whether the specified budget has been exhausted already.
stream.budgetmanager.VariableUncertaintyBudgetManager	Budget manager which checks, whether the specified budget has been exhausted already.
stream.budgetmanager.SplitBudgetManager	Budget manager which checks, whether the specified budget has been exhausted already.
stream.budgetmanager.BalancedIncrementalQuantileFilter	The Balanced Incremental Quantile Filter has been proposed together with Probabilistic Active Learning for Datastreams [1].
stream.budgetmanager.RandomVariableUncertaintyBudgetManager	Budget manager which checks, whether the specified budget has been exhausted already.
stream.budgetmanager.DensityBasedSplitBudgetManager	Budget manager which checks, whether the specified budget has been exhausted already.
stream.budgetmanager.RandomBudgetManager	Budget manager which checks, whether the specified budget has been exhausted already.

skactiveml.classifier:

The skactiveml.classifier module.

Classes

classifier.ParzenWindowClassifier	The Parzen window classifier (PWC) is a simple and probabilistic classifier.
classifier.MixtureModelClassifier	The classifier based on a mixture model (MixtureModelClassifier) is a generative classifier based on a (Bayesian) Gaussian mixture model (GMM).
classifier.SklearnClassifier	Implementation of a wrapper class for scikit-learn classifiers such that missing labels can be handled.
classifier.SlidingWindowClassifier	Implementation of a wrapper class for SkactivemlClassifier such that the number of training samples can be limited to the latest window_size samples.

skactiveml.classifier.multiannotator:

Classes

classifier.multiannotator.AnnotatorLogisticRegression	Logistic Regression based on Raykar [1] is a classification algorithm that learns from multiple annotators.
classifier.multiannotator.AnnotatorEnsembleClassifier	This strategy consists of fitting one classifier per annotator.

skactiveml.visualization:

The skactiveml.visualization module includes various tools for visualization.

Functions

visualization.plot_utilities	Plot the utility for the given single-annotator query strategy.
visualization.plot_decision_boundary	Plot the decision boundary of the given classifier.
visualization.plot_contour_for_samples	Plot the utility for the given query strategy.
visualization.plot_annotator_utilities	Plot the utility for the given query strategy.
visualization.plot_stream_training_data	Plot the utility for the given query strategy.
visualization.plot_stream_decision_boundary	Plot the decision boundary of the given classifier.
visualization.mesh	Function to get instances of a mesh grid as well as x-mesh and y-mesh with given resolution in the specified bounds.

skactiveml.utils:

The skactiveml.utils module includes various utilities.

Classes

utils.ExtLabelEncoder

Encode class labels with value between 0 and classes-1 and uses -1 for unlabeled samples.

Functions

utils.rand_argmax	Returns index of maximum value.
utils.rand_argmin	Returns index of minimum value.
utils.compute_vote_vectors	Counts number of votes per class label for each sample.
utils.majority_vote	Assigns a label to each sample based on weighted voting.
utils.is_unlabeled	Creates a boolean mask indicating missing labels.
utils.is_labeled	Creates a boolean mask indicating present labels.
utils.check_classes	Check whether class labels are uniformly strings or numbers.
utils.check_missing_label	Check whether a missing label is compatible to a given target type.
utils.check_cost_matrix	Check whether cost matrix has shape (n_classes, n_classes).
utils.check_scalar	Validate scalar parameters type and value.
utils.check_classifier_params	Check whether the parameters are compatible to each other (only if classes is not None).
utils.check_X_y	Input validation for standard estimators.
utils.check_random_state	Check validity of the given random state.
utils.call_func	Calls a function with the given parameters given in kwargs if they exist as parameters in f_callable.
utils.match_signature	A decorator that matches the signature to a given method from a reference and hides it when the reference object does not have the wrapped function.
utils.check_class_prior	Check if the class_prior is a valid prior.
utils.ext_confusion_matrix	Compute confusion matrix to evaluate the accuracy of a classification.
utils.labeled_indices	Return an array of indices indicating present labels.
utils.unlabeled_indices	Return an array of indices indicating missing labels.
utils.check_type	Check if obj is one of the given types.
utils.check_bound	Validates bound and returns the bound of X if bound is None.
utils.check_equal_missing_label	Check whether two missing label values are equal to each other.
utils.check_budget_manager	Validate if budget manager is a budgetmanager class and create a copy 'budget_manager_'.
utils.check_indices	Check if indices fit to array.
utils.simple_batch	Generates a batch by selecting the highest values in the 'utilities'.
utils._check_callable	Checks if function is a callable and if the number of free parameters is correct.

skactiveml.exceptions:

Classes

exceptions.MappingError

Exception class to raise if a strategy needs a mapping between samples and candidates which is not available.