filter_by_confidence#

movement.filtering.filter_by_confidence(data, confidence, threshold=0.6, keep_points_with_nan_confidence=True, print_report=False)[source]#

Drop data points below a certain confidence threshold.

Data points with an associated confidence value below the threshold are converted to NaN. Data points with a missing (NaN) confidence value are kept by default, but this can be changed via keep_points_with_nan_confidence.

Parameters:

  • data (DataArray) – The input data to be filtered.

  • confidence (DataArray) – The data array containing confidence scores to filter by.

  • threshold (float) – The confidence threshold below which datapoints are filtered. A default value of 0.6 is used. See notes for more information.

  • keep_points_with_nan_confidence (bool) – Whether to keep data points whose confidence value is NaN. Default is True. See notes for more information.

  • print_report (bool) – Whether to print a report on the number of NaNs in the dataset before and after filtering. Default is False.

Returns:

The data where points with a confidence value below the user-defined threshold have been converted to NaNs.

Return type:

DataArray

Notes

For the poses dataset case, note that the point-wise confidence values reported by various pose estimation frameworks are not standardised, and the range of values can vary. For example, DeepLabCut reports a likelihood value between 0 and 1, whereas the point confidence reported by SLEAP can range above 1. Therefore, the default threshold value will not be appropriate for all datasets and does not have the same meaning across pose estimation frameworks. We advise users to inspect the confidence values in their dataset and adjust the threshold accordingly.

A missing (NaN) confidence value may mean that the point was manually annotated or edited, and therefore has no associated model confidence. For example, this is the case for points proof-read in SLEAP. Such points are the ones we should be most confident about, so they are kept by default. Set keep_points_with_nan_confidence=False to drop them instead.

Examples using filter_by_confidence#

Annotate and load events with BORIS

Annotate and load events with BORIS

Drop outliers and interpolate

Drop outliers and interpolate

Scale pose tracks to real-world units

Scale pose tracks to real-world units