sleap.info.metrics#

Module for producing prediction metrics for SLEAP datasets.

sleap.info.metrics.calculate_pairwise_cost(instances_a: List[Union[Instance, PredictedInstance]], instances_b: List[Union[Instance, PredictedInstance]], cost_function: Callable) → ndarray[source]#

Calculate (a * b) matrix of pairwise costs using cost function.

sleap.info.metrics.compare_instance_lists(instances_a: List[Union[Instance, PredictedInstance]], instances_b: List[Union[Instance, PredictedInstance]]) → ndarray[source]#

Given two lists of corresponding Instances, returns (instances * nodes) matrix of distances between corresponding nodes.

sleap.info.metrics.list_points_array(instances: List[Union[Instance, PredictedInstance]]) → ndarray[source]#

Given list of Instances, returns (instances * nodes * 2) matrix.

sleap.info.metrics.match_instance_lists(instances_a: List[Union[Instance, PredictedInstance]], instances_b: List[Union[Instance, PredictedInstance]], cost_function: Callable) → Tuple[List[Union[Instance, PredictedInstance]], List[Union[Instance, PredictedInstance]]][source]#

Sorts two lists of Instances to find best overall correspondence for a given cost function (e.g., total distance between points).

sleap.info.metrics.match_instance_lists_nodewise(instances_a: List[Union[Instance, PredictedInstance]], instances_b: List[Union[Instance, PredictedInstance]], thresh: float = 5) → Tuple[List[Union[Instance, PredictedInstance]], List[Union[Instance, PredictedInstance]]][source]#

For each node for each instance in the first list, pairs it with the closest corresponding node from any instance in the second list.

sleap.info.metrics.matched_instance_distances(labels_gt: ~sleap.io.dataset.Labels, labels_pr: ~sleap.io.dataset.Labels, match_lists_function: ~typing.Callable = <function match_instance_lists_nodewise>, frame_range: ~typing.Optional[range] = None) → Tuple[List[int], ndarray, ndarray, ndarray][source]#

Distances between ground truth and predicted nodes over a set of frames.

Parameters:

  • labels_gt – the Labels object with ground truth data

  • labels_pr – the Labels object with predicted data

  • match_lists_function – function for determining corresponding instances Takes two lists of instances and returns “sorted” lists.

  • frame_range (optional) – range of frames for which to compare data If None, we compare every frame in labels_gt with corresponding frame in labels_pr.

Returns:

  • frame indices map: instance idx (for other matrices) -> frame idx

  • distance matrix: (instances * nodes)

  • ground truth points matrix: (instances * nodes * 2)

  • predicted points matrix: (instances * nodes * 2)

Return type:

Tuple

sleap.info.metrics.nodeless_point_dist(inst_a: Union[Instance, PredictedInstance], inst_b: Union[Instance, PredictedInstance]) → ndarray[source]#

Given two instances, returns array of distances for closest points ignoring node identities.

sleap.info.metrics.point_dist(inst_a: Union[Instance, PredictedInstance], inst_b: Union[Instance, PredictedInstance]) → ndarray[source]#

Given two instances, returns array of distances for corresponding nodes.

sleap.info.metrics.point_match_count(dist_array: ndarray, thresh: float = 5) → int[source]#

Given an array of distances, returns number which are <= threshold.

sleap.info.metrics.point_nonmatch_count(dist_array: ndarray, thresh: float = 5) → int[source]#

Given an array of distances, returns number which are not <= threshold.