Lightning Components

class InnerEye.ML.lightning_metrics.Accuracy05(*args: Any, **kwargs: Any)

correct: Tensor

property has_predictions: bool

Returns True if the present object stores at least 1 prediction (self.update has been called at least once), or False if no predictions are stored.

total: Tensor

class InnerEye.ML.lightning_metrics.AccuracyAtOptimalThreshold

Computes the binary classification accuracy at an optimal cut-off point.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.AreaUnderPrecisionRecallCurve

Computes the area under the precision-recall-curve.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.AreaUnderRocCurve

Computes the area under the receiver operating curve (ROC).

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.AverageWithoutNan(dist_sync_on_step: bool = False, name: str = '')

A generic metric computer that keep track of the average of all values excluding those that are NaN.

compute() → Tensor

Override this method to compute the final metric value from state variables synchronized across the distributed backend.

update(value: Tensor) → None

Stores all the given individual elements of the given tensor in the present object.

class InnerEye.ML.lightning_metrics.BinaryCrossEntropyWithLogits

Computes the cross entropy for binary classification. This metric must be computed off the model output logits.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.ExplainedVariance(*args: Any, **kwargs: Any)

property has_predictions: bool

Returns True if the present object stores at least 1 prediction (self.update has been called at least once), or False if no predictions are stored.

n_obs: Tensor

sum_error: Tensor

sum_squared_error: Tensor

sum_squared_target: Tensor

sum_target: Tensor

class InnerEye.ML.lightning_metrics.FalseNegativeRateOptimalThreshold

Computes the false negative rate when choosing the optimal cut-off point for a binary classifier.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.FalsePositiveRateOptimalThreshold

Computes the false positive rate when choosing the optimal cut-off point for a binary classifier.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.MeanAbsoluteError(*args: Any, **kwargs: Any)

property has_predictions: bool

Returns True if the present object stores at least 1 prediction (self.update has been called at least once), or False if no predictions are stored.

sum_abs_error: Tensor

total: Tensor

class InnerEye.ML.lightning_metrics.MeanSquaredError(*args: Any, **kwargs: Any)

property has_predictions: bool

Returns True if the present object stores at least 1 prediction (self.update has been called at least once), or False if no predictions are stored.

sum_squared_error: Tensor

total: Tensor

class InnerEye.ML.lightning_metrics.MetricForMultipleStructures(ground_truth_ids: List[str], is_training: bool, metric_name: str = 'Dice', use_average_across_structures: bool = True)

Stores a metric for multiple structures, and an average Dice score across all structures. The class consumes pre-computed metric values, and only keeps an aggregate for later computing the averages. When averaging, metric values that are NaN are skipped.

compute_all() → Iterator[Tuple[str, Tensor]]

Calls the .compute() method on all the metrics that the present object holds, and returns a sequence of (metric name, metric value) tuples. This will automatically also call .reset() on the metrics. The first returned metric is the average across all structures, then come the per-structure values.

reset() → None

Calls the .reset() method on all the metrics that the present object holds.

training: bool

update(values_per_structure: Tensor) → None

Stores a vector of per-structure Dice scores in the present object. It updates the per-structure values, and the aggregate value across all structures.

Parameters:

values_per_structure – A row tensor that has as many entries as there are ground truth IDs.

class InnerEye.ML.lightning_metrics.OptimalThreshold

Computes the optimal cut-off point for a binary classifier.

compute() → Tensor

Computes a metric from the stored predictions and targets.

class InnerEye.ML.lightning_metrics.ScalarMetricsBase(name: str = '', compute_from_logits: bool = False)

A base class for all metrics that can only be computed once the complete set of model predictions and labels is available. The base class provides an update method, and synchronized storage for predictions (field preds) and labels (field targets). Derived classes need to override the compute method.

compute() → Tensor

Computes a metric from the stored predictions and targets.

property has_predictions: bool

Returns True if the present object stores at least 1 prediction (self.update has been called at least once), or False if no predictions are stored.

update(preds: Tensor, targets: Tensor) → None

Override this method to update the state variables of your metric class.

InnerEye.ML.lightning_metrics.nanmean(values: Tensor) → Tensor

Computes the average of all values in the tensor, skipping those entries that are NaN (not a number). If all values are NaN, the result is also NaN.

Parameters:

values – The values to average.

Returns:

A scalar tensor containing the average.