metric

Metric

Bases: Named, Protocol

Minimal template for metrics.

Call flow

call -> prepare(true), prepare(predicted) -> compute(true, predicted)

prepare(data)

Hook to normalize/collect/select data before computing metric.

Default: identity. Mixins override and call super().prepare(...)

Source code in src/flowcean/core/metric.py

def prepare(self, data: Data) -> Data:
    """Hook to normalize/collect/select data before computing metric.

    Default: identity. Mixins override and call super().prepare(...)
    """
    return data

__call__(true, predicted)

Execute metric: prepare inputs then compute.

Source code in src/flowcean/core/metric.py

@final
def __call__(
    self,
    true: Data,
    predicted: Data,
) -> Reportable | dict[str, Reportable]:
    """Execute metric: prepare inputs then compute."""
    return self.compute(true, predicted)

compute(true, predicted)

Implement metric logic on prepared inputs.

Source code in src/flowcean/core/metric.py

def compute(self, true: Data, predicted: Data) -> Reportable:
    """Implement metric logic on prepared inputs."""
    t = self.prepare(true)
    p = self.prepare(predicted)
    return self._compute(t, p)

ActiveMetric

Bases: Named, Protocol

Base class for metrics for active environments.

__call__(observations, actions) abstractmethod

Calculate the metric value based on the observations.

Parameters:

Name	Type	Description	Default
observations	list[Observation]	list of observations of the environment	required
actions	list[Action]	list of actions of the learner	required

Returns:

Type	Description
Reportable | dict[str, Reportable]	Metric value

Source code in src/flowcean/core/metric.py

@abstractmethod
def __call__(
    self,
    observations: list[Observation],
    actions: list[Action],
) -> Reportable | dict[str, Reportable]:
    """Calculate the metric value based on the observations.

    Args:
        observations: list of observations of the environment
        actions: list of actions of the learner

    Returns:
        Metric value
    """