climb.tool.impl.data_suite.third_party.uq360.utils.transformers package

Subpackages

• climb.tool.impl.data_suite.third_party.uq360.utils.transformers.nearest_neighbors package
  • Submodules
  • climb.tool.impl.data_suite.third_party.uq360.utils.transformers.nearest_neighbors.base module
  • climb.tool.impl.data_suite.third_party.uq360.utils.transformers.nearest_neighbors.exact module
  • climb.tool.impl.data_suite.third_party.uq360.utils.transformers.nearest_neighbors.faiss module
  • climb.tool.impl.data_suite.third_party.uq360.utils.transformers.nearest_neighbors.pynndescent module
  • Module contents

Submodules

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.class_accuracy module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.class_accuracy.ClassAccuracyTransformer(model=None)

Bases: FeatureTransformer

Test set accuracy of the input/baseline model for samples in the predicted class.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.class_frequency module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.class_frequency.ClassFrequencyTransformer

Bases: FeatureTransformer

Fraction of the train set belonging to the predicted class.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.cluster_feature module

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_delta module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_delta.ConfidenceDeltaTransformer

Bases: FeatureTransformer

Highest minus second highest class confidence from the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_entropy module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_entropy.ConfidenceEntropyTransformer

Bases: FeatureTransformer

Entropy of the confidence vector from the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_min_max module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_min_max.ConfidenceMinMaxTransformer

Bases: FeatureTransformer

Ratio of the minimum and maximum class confidences from the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_std module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_std.ConfidenceStdTransformer

Bases: FeatureTransformer

Standard deviation of the confidence vector from the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_top module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.confidence_top.ConfidenceTopTransformer

Bases: FeatureTransformer

Highest class confidence from the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.distribution_clustering module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.distribution_clustering.DistributionClusteringTransformer(scaling_exponent=6, min_cluster_points=12)

Bases: FeatureTransformer

HDBScan clustering based feature transformer.

At fit time, this transformer just fits a standard-scaling to the training data. At predict time, it uses hdbscan to cluster the production data.

For efficiency, the data used in the clustering is randomly downsampled to 30,000 data points.

WARNING: If this happens, the output will not have the same first axis size as the input.

The output at inference time is the centroid position in feature space of the cluster that each inference point belongs to, concatenated with the proportion of total points that were contained in that cluster.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

rescale(X)

save(output_location=None)

set_feature_importances(feature_importances)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.feature_transformer module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.feature_transformer.FeatureTransformer

Bases: Base

Base class for feature transformers: derived features based on the feature vectors of the input dataset. Given an input array of shape (N_samples, M_features), they output an array of shape (N_samples, P_derived_features).

fit(x, y)

classmethod instance(subtype_name=None, **params)

property json_registry

load(input_location=None)

property pkl_registry

register_json_object(obj, name)

register_pkl_object(obj, name)

save(output_location=None)

transform()

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.gbm module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.gbm.GBMTransformer

Bases: FeatureTransformer

GBM shadow-model feature. This class trains a GBM model on the same train set as the input/baseline model. At inference time, the top class confidence and top - 2nd class confidence are used as the derived feature.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.group_scaler module

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.logistic_regression module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.logistic_regression.LogisticRegressionTransformer(model=None)

Bases: FeatureTransformer

Logistic regression shadow-model feature. This class trains a GBM model on the same train set as the input/baseline model. At inference time, the top class confidence and top - 2nd class confidence are used as the derived feature.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.mlp module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.mlp.MLPTransformer

Bases: FeatureTransformer

Four layer mlp shadow-model feature. This class trains a GBM model on the same train set as the input/baseline model. At inference time, the top class confidence and top - 2nd class confidence are used as the derived feature.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.one_class_svm module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.one_class_svm.OneClassSVMTransformer

Bases: FeatureTransformer

One-class SVM outlier-classifier based derived feature. This transformer fits an SVM decision boundary enclosing the full training set. This is then the decision boundary to identify outliers in production data at inference time.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.original_features module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.original_features.OriginalFeaturesTransformer

Bases: FeatureTransformer

Dummy/identity transformer which passes the data array through unchanged.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.pca module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.pca.PCATransformer(k=2)

Bases: FeatureTransformer

Transformer which applies a standard scaling followed by a PCA decomposition to the dataset, then returns only the k data components with the highest variance (ie the first k in the PCA decomposition).

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.predicted_class module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.predicted_class.PredictedClassTransformer

Bases: FeatureTransformer

Derived feature which uses the label (integer) of the class predicted by the input/base model.

load(input_dir=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_dir=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.random_forest module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.random_forest.RandomForestTransformer

Bases: FeatureTransformer

Random forest shadow-model feature. This class trains a GBM model on the same train set as the input/baseline model. At inference time, the top class confidence and top - 2nd class confidence are used as the derived feature.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.svc module

class climb.tool.impl.data_suite.third_party.uq360.utils.transformers.svc.SVCTransformer

Bases: FeatureTransformer

Support-vector machine shadow-model feature. This class trains a GBM model on the same train set as the input/baseline model. At inference time, the top class confidence and top - 2nd class confidence are used as the derived feature.

fit(x, y)

load(input_location=None)

classmethod name()

Name of this subtype, used for lookup purposes. To be implemented by subclasses. This method can either return a single name, or a list/tuple of names.

save(output_location=None)

transform(x, predictions)

climb.tool.impl.data_suite.third_party.uq360.utils.transformers.umap_kde module

Module contents