tempor.metrics.metric module¶

Module containing the base class for metrics.

class tempor.metrics.metric.Metric[source]¶

Bases: Plugin, ABC

Metric abstract base class, defines the required methods.

The base class that all plugins must inherit from.

abstract property direction : Literal[minimize] | Literal[maximize]¶: The direction of the metric

evaluate(actual: Any, predicted: Any, *args: Any, **kwargs: Any) → Any[source]¶

The metric evaluation call.

Parameters:¶

actual : Any¶: Actual value(s).
predicted : Any¶: Predicted value(s).
*args : Any: Additional positional arguments.
**kwargs : Any: Additional keyword arguments.

Returns:¶

Evaluated metric value(s)

Return type:¶

Any

class tempor.metrics.metric.OneOffPredictionMetric[source]¶

Bases: Metric

Metric abstract base class for the one-off prediction task.

The base class that all plugins must inherit from.

class tempor.metrics.metric.OneOffClassificationMetric[source]¶

Bases: OneOffPredictionMetric

Metric abstract base class for the one-off prediction task, classification case.

The base class that all plugins must inherit from.

evaluate(actual: ndarray, predicted: ndarray, *args: Any, **kwargs: Any) → float[source]¶

The metric evaluation call.

actual and predicted are expected to be numpy arrays (sample in the 0th dimension).

predicted must be the probabilities in this case.

Parameters:¶

actual : np.ndarray¶: Actual value(s).
predicted : np.ndarray¶: Predicted value(s).
*args : Any: Additional positional arguments.
**kwargs : Any: Additional keyword arguments.

Returns:¶

Evaluated metric value.

Return type:¶

float

class tempor.metrics.metric.OneOffRegressionMetric[source]¶

Bases: OneOffPredictionMetric

Metric abstract base class for the one-off prediction task, regression case.

The base class that all plugins must inherit from.

evaluate(actual: ndarray, predicted: ndarray, *args: Any, **kwargs: Any) → float[source]¶

The metric evaluation call.

actual and predicted are expected to be numpy arrays (sample in the 0th dimension).

Parameters:¶

actual : np.ndarray¶: Actual value(s).
predicted : np.ndarray¶: Predicted value(s).
*args : Any: Additional positional arguments.
**kwargs : Any: Additional keyword arguments.

Returns:¶

Evaluated metric value.

Return type:¶

float

class tempor.metrics.metric.TimeToEventMetric[source]¶

Bases: Metric

Metric abstract base class for the time-to-event (survival) task.

The base class that all plugins must inherit from.

evaluate(actual: tuple[ndarray, ndarray], predicted: ndarray, horizons: list[float] | list[int] | list[Timestamp], *args: Any, **kwargs: Any) → list[float][source]¶

The metric evaluation call.

Parameters:¶

actual : metric_typing.EventArrayTimeArray¶: A tuple of two numpy arrays: the event values array and the event times array, for the actual event vales.
predicted : np.ndarray¶: A numpy array of shape (n_samples, n_horizons_timesteps, n_features) with the predicted risk estimates.
horizons : data_typing.TimeIndex¶: List of horizons time points.
*args : Any: Additional positional arguments.
**kwargs : Any: Additional keyword arguments.

Returns:¶

The metric values for each horizon time point.

Return type:¶

List[float]