quantus.helpers.enums module

class quantus.helpers.enums.DataType(value)

Bases: Enum

This enum represents the different types of data that a metric implementation currently supports.

  • IMAGE: Represents image data.

  • TABULAR: Represents tabular data.

  • TEXT: Represents text data.

IMAGE = 'image'
TABULAR = 'tabular'
TEXT = 'text'
TIMESERIES = 'time-series'
class quantus.helpers.enums.EvaluationCategory(value)

Bases: Enum

This enum represents different categories of explanation quality for XAI algorithms.

  • FAITHFULNESS: Indicates how well the explanation reflects the true features used by the model.

  • ROBUSTNESS: Represents the degree to which the explanation remains consistent under small perturbations in the input.

  • RANDOMISATION: Measures the quality of the explanation in terms of difference in explanation when randomness is introduced.

  • COMPLEXITY: Refers to how easy it is to understand the explanation. Lower complexity is usually better.

  • LOCALISATION: Refers to how consistently the explanation points out the parts of the input as defined in a ground-truth segmentation mask.

  • AXIOMATIC: Represents the quality of the explanation in terms of well-defined axioms.

AXIOMATIC = 'Axiomatic'
COMPLEXITY = 'Complexity'
FAITHFULNESS = 'Faithfulness'
LOCALISATION = 'Localisation'
NONE = 'None'
RANDOMISATION = 'Randomisation'
ROBUSTNESS = 'Robustness'
class quantus.helpers.enums.ModelType(value)

Bases: Enum

This enum represents the different types of models that a metric can work with.

  • TORCH: Represents PyTorch models.

  • TF: Represents TensorFlow models.

TF = 'tensorflow'
TORCH = 'torch'
class quantus.helpers.enums.ScoreDirection(value)

Bases: Enum

This enum represents the direction that the score of a metric should go in for better results.

  • HIGHER: Higher scores are better.

  • LOWER: Lower scores are better.

HIGHER = 'higher'
LOWER = 'lower'