statistic

Handle statistics.

Classes:

Name	Description
Stat	Represents a statistic for the explainable model.
HistogramStat	Histogram statistics metric.
VarianceStat	Histogram statistics metric.
DistributionStat	Distribution statistics metric.
DictStats	Dictionary of statistics (torch eval etc...) used in training or explain.

Stat

Represents a statistic for the explainable model.

HistogramStat

Histogram statistics metric.

Parameters:

Name	Type	Description	Default
on	Literal['input', 'target', 'prediction', 'prediction_error']	Data type to compute the statistic on.	required
num_bins	int	Number of bins linearly spaced to compute from the sampled items.	required
num_items	int	Number total of items to sample from the data to approximate a histogram, as not all samples are used for memory efficiency.	required
on_raw_data	bool	Whether to compute the statistic on the raw spce or on the preprocessed space.	required
feature_name	str | None	In case of "input" data type, the statistic will be computed only for the feature named feature_name. If None, it will compute the statistic on the schema first feature.	None

Attributes:

Name	Type	Description
on	Literal['input', 'target', 'prediction', 'prediction_error']
num_bins	int
num_items	int
on_raw_data	bool
feature_name	str | None

on: Literal['input', 'target', 'prediction', 'prediction_error']

num_bins: int

num_items: int

on_raw_data: bool

feature_name: str | None = None

VarianceStat

Histogram statistics metric.

Parameters:

Name	Type	Description	Default
on	Literal['input', 'target', 'prediction', 'prediction_error']	Data type to compute the statistic on.	required
on_raw_data	bool	Whether to compute the statistic on the raw spce or on the preprocessed space.	required
feature_name	str | None	In case of "input" data type, the statistic will be computed only for the feature named feature_name. If None, it will compute the statistic on the schema first feature.	None

Attributes:

Name	Type	Description
on	Literal['input', 'target', 'prediction', 'prediction_error']
on_raw_data	bool
feature_name	str | None

on: Literal['input', 'target', 'prediction', 'prediction_error']

on_raw_data: bool

feature_name: str | None = None

DistributionStat

Distribution statistics metric.

The distribution is always computed on the preprocessed space as it requires one-hot vectors for the categorical feature.

Parameters:

Name	Type	Description	Default
on	Literal['input', 'target', 'prediction', 'prediction_error']	Data type to compute the statistic on.	required
feature_name	str | None	In case of "input" data type, the statistic will be computed only for the feature named feature_name. If None, it will compute the statistic on the schema first feature.	None

Attributes:

Name	Type	Description
on	Literal['input', 'target', 'prediction', 'prediction_error']
feature_name	str | None

on: Literal['input', 'target', 'prediction', 'prediction_error']

feature_name: str | None = None

DictStats

Dictionary of statistics (torch eval etc...) used in training or explain.

Each stat will be computed globally (for the whole model) and per leaf.

Stats are displayed in XpViz along with explanations. Dictionary keys are will be used as metric names in XpViz.