Is there a way to exclude specific syllables that don't match a behavior otherwise represented in a syllable?

[sageleland]

I am working on a project classifying the behavior of mice during a social interaction test after chronic social defeat stress. After I created the syllables, it looks to me like about 70% of a given syllable when generating the video visualization displays a certain behavior such as rearing. Is there a way I can exclude the ones that aren't? Thank you!

[JRJacoby]

The grid videos display only 20 random instances of each syllable, which is usually a very small proportion of all of the instances. Changing a few of the ones that you can see would likely have little to no effect on the overall results.

If you are concerned that this is a widespread issue, one thing you could try is running the AR_only model for a full 500 iterations. This sometimes pulls apart similar syllables more robustly. However, the advantage of the full model is that it handles erroneous keypoint jitter. So if your data has a lot of keypoint jitter, the AR_only model will pick up that jitter as a syllable.