Changelog

2.1.5

New Features

New tutorial for the AnDi 2 Benchmark dataset (#49)
New analysis tools: p-variation and Power spectral density as well as the Cramer-Rao lower bounds (#50)
Local version of the scoring program used during the AnDi 2 competition that allows for local scoring ((#51)

Bugs Squashed

Solved bug for indexing VIP particles in videos ( #44 and #46)
Recoded label_filter to solve many existing bugs (#40)
Solved bug in slicing in missing trajectories (see #39)
Improved utils_challenge.label_filter and added few extra tests to ensure that there are never trajectories with segments shorter than min_seg.

2.1.4

New Features

Pathified utils_challenge.file_nonOverlap_reOrg (see #32 and #37): now paths are pathlib.Path rather than strings.
Included trap radius as input to utils_trajectories.plot_trajs (see #33).
Improved the message when no trajectories are found in the FOV (see #36).

Bugs Squashed

Improved utils_challenge.label_filter and introduced utils_challenge.unique_labelled to properly handed filter labelling and avoid segments smaller than min_seg. The issue arised because the previous labelling for the utils_challenge.majority_filter would vary depending on the actual values in the input array, creating labels based on the sort vector rather than in order of appearance.
Corrected loop over self.dics in datasets_phenom.create_dataset (see #35).

2.1.3

(not released in Pypi)

New Features

ensemble_changepoint_error: improved the metric such that cases in which groundtruth (GT) has not changepoint (CP) but prediction (pred) has few trajectories with CP does not give maximum error. We consider now that each no GT trajectory correctly predicted is a true positive. For the TP_RMSE, we have changed the case in which, if you had no CP, TP_rmse would be maximal. To be closer to the AnDi2 definition, if there were no TP, we set TP_RMSE = 0.