Short Trajectories

[ryanredward]

Is there any way to manually label trajectories that are too short to be segmented? Currently, I have not been able to label any direct findings because of this issue.

Please advise. Thanks!

[avouros]

Hi,

All of the unsegmented trajectories are automatically assigned to Direct Finding. You can change this by manually labelling them during the results generation step

DF and AT is used (usually) only on the full trajectories (this is why they have 'trajectory' as a note). To label full trajectories load my_trajectories.mat inside the settings subfolder of your project folder.

The above applies on version 4 of the software which will be officially out during weekend (still working on some bugs). It will be announced here.

[avouros]

Hi ryanredward new release (v4.0.2) is out! Check it out and please let us know if you come across any issues.

[lolaBerkowitz]

Is there anyway to create a training set based on full trajectories(that are shorter than the segment length)? Not all paths shorter than the specified segment length are direct trajectories or approaching target. This is particularly troublesome with my data set.

Looking forward to your response! Thanks!

[avouros]

Unfortunately no, full trajectories should all be labelled manually. I have put this limitation because our methodology has been tested only on segmented trajectories. Some limited tested on full trajectory classification has been performed but it yields poor classification results due to the fact that full trajectories usually fall under many categories.

However, you can label unsegmented trajectories during the generation of your results (see https://github.com/RodentDataAnalytics/mwm-ml-gen/wiki/Results#EXTRAS). Then these trajectories will be considered as segments and be analysed along with the rest of your trajectory segments.
(Currently this option is not very user friendly because each time you have to relabel these trajectories, I will try to fix that in the future.)

[lolaBerkowitz]

Got it thanks!