quantus.helpers.warn module

This modules holds a collection of perturbation functions i.e., ways to perturb an input or an explanation.

quantus.helpers.warn.check_kwargs(kwargs)

Check that no additional kwargs are passed, i.e. the kwargs dict is empty. Raises an exception with helpful suggestions to fix the issue.

Parameters:
kwargs: optional

Keyword arguments.

Returns:
None
quantus.helpers.warn.deprecation_warnings(kwargs: dict) None

Run deprecation warnings.

Parameters:
kwargs: optional

Keyword arguments.

Returns:
None
quantus.helpers.warn.warn_absolute_operation(word: str = '') None

Warn if an absolute operation is applied, where the metric is defined otherwise.

Parameters:
word: string

A string for which is ‘’ or ‘not ‘.

Returns:
None
quantus.helpers.warn.warn_different_array_lengths() None

Warn if the array lengths are different, for plotting.

Returns:
None
quantus.helpers.warn.warn_empty_segmentation() None

Warn if the segmentation mask is empty.

Returns:
None
quantus.helpers.warn.warn_iterations_exceed_patch_number(n_iterations: int, n_patches: int) None

Warn if the number of non-overlapping patches is lower than the number of iterations specified for this metric.

Parameters:
n_iterations: integer

The number of iterations specified in the metric.

n_patches: integer

The number of patches specified in the metric.

Returns:
None
quantus.helpers.warn.warn_max_size() None

Warns if the ratio is smaller than the maximum size, for attribution_localisaiton metric. Returns ——- None

quantus.helpers.warn.warn_noise_zero(noise: float) None

Warn if noise is zero.

Parameters:
noise: float

The amount of noise.

Returns:
None
quantus.helpers.warn.warn_normalise_operation(word: str = '') None

Warn if a normalisation operation is applied, where the metric is defined otherwise.

Parameters:
word: string

A string for which is ‘’ or ‘not ‘.

Returns:
None
quantus.helpers.warn.warn_parameterisation(metric_name: str = 'Metric', sensitive_params: str = 'X, Y and Z.', data_domain_applicability: str = '', citation: str = 'INSERT CITATION')

Warn the parameterisation of the metric.

Parameters:
metric_name: string

The metric name.

sensitive_params: string

The sensitive parameters of the metric.

data_domain_applicability string

The applicability when it comes to data domains, default = “”.

citation: string

The citation.

Returns:
None
quantus.helpers.warn.warn_perturbation_caused_no_change(x: ndarray, x_perturbed: ndarray) None

Warn that perturbation applied to input caused no change so that input and perturbed input is the same.

Parameters:
x: np.ndarray

The original input that is considered unperturbed.

x_perturbed: np.ndarray

The perturbed input.

Returns:
None
quantus.helpers.warn.warn_segmentation(inside_attribution: float, total_attribution: float) None

Warn if the inside explanation is greater than total explanation.

Parameters:
inside_attribution: float

The size of inside attribution.

total_attribution: float

The size of total attribution.

Returns:
None