yujiahu415
commented
8 months ago Hi,
-
The metrics of training and testing may be different because they are done on different samples. The more samples used for calculating these metrics when doing training and testing, the more consistent of the metrics between training and testing. The supporting number is the sample number and the f1 score is the weighted average of precision and recall (sensitivity).
-
Yes, the Categorizer should be able to generalize on different videos with different mice.
-
The error indicates your system ran out of memory. When the Categorizer does the behavioral classification, it requires additional memory. Since the duration of your video to analyze is pretty long, 25GB might not be enough. You may either increase the system memory or trim the video into shorter clips.
vzimmern
Good afternoon,
Three questions about this excellent software:
How should I interpret the table that emerges when training and testing categorizers? Specifically, the table reports that one of my behaviors has a precision of 0.37 after the training process and that number changes to 0.5 when testing the categorizers. How should I interpret these numbers? How should I interpret the f1-score and the support score?
I am trying to identify a rare behavior in a mouse. For this rare behavior, can I combine behaviors from multiple different videos (with different mice) to get a higher number of behaviors for the categorizer to train on?
I have created a categorizer and am evaluating an hour of video recording of a single mouse. The process seems to die whenever the categorizer starts being used. The last time the process was killed it was because of lack of memory -- but this time, I have checked the memory and there's plenty of RAM left (~ 25 GB left) while SWAP memory is maxed out at 2 GB.
Here's a screenshot of the terminal output preceding the process kill.
Thanks again for the help!