familiarNoveltyDetector-class: Novelty detector.
In familiar: End-to-End Automated Machine Learning and Model Evaluation

familiarNoveltyDetector-class

R Documentation

Novelty detector.

Description

A familiarNoveltyDetector object is a self-contained model that can be applied to generate out-of-distribution predictions for instances in a dataset.

Details

Note that these objects do not contain any data concerning outcome, as this not relevant for (prospective) out-of-distribution detection.

Slots

name: Name of the familiarNoveltyDetector object.
model: The actual novelty detector trained using a specific algorithm, e.g. a isolation forest from the isotree package.
feature_info: List of objects containing feature information, e.g., name, class levels, transformation, normalisation and clustering parameters.
data_column_info: Data information object containing information regarding identifier column names.
hyperparameters: Set of hyperparameters used to train the detector.
hyperparameter_data: Information generated during hyperparameter optimisation. Currently not used.
calibration_model: Model used to convert raw output to statistical probability of being out-of-distribution. Currently not used.
learner: Learning algorithm used to create the novelty detector.
vimp_method: Method used to determine variable importance for the novelty detector.
vimp_table: Variable importance table or list of variable importance tables for the model.
vimp_aggregation_method: Method used for aggregating variable importance tables if more than one is present.)
vimp_rank_threshold: Threshold used for some variable importance aggregation methods.
required_features: The set of features required for complete reproduction, i.e. with imputation.
model_features: The set of features that is used to train the detector.
data_id: Internal identifier for the dataset used to train the detector.
run_id: Internal identifier for the specific subset of the dataset used used to train the detector.
run_table: Run table for the data used to train the detector. Used internally.
is_trimmed: Flag that indicates whether the detector, stored in the model slot, has been trimmed.
trimmed_function: List of functions whose output has been captured prior to trimming the model.
messages: List of warning and error messages generated during training.
project_id: Identifier of the project that generated the familiarNoveltyDetector object.
familiar_version: Version of the familiar package.
package: Name of package(s) required to executed the detector itself, e.g. isotree.
package_version: Version of the packages mentioned in the package attribute.