search

dattalab / keypoint-moseq

https://keypoint-moseq.readthedocs.io
Other
64 stars 26 forks source link

Estimated bodyparts missing from kpms 2.0 #81

Closed ellaksay closed 11 months ago

ellaksay commented 11 months ago

Hello, I recently updated to 2.0 and ran my old data generated with 1.5 through the re-indexing pipeline. I was wondering if there was a way for the code to retain the estimated bodyparts columns that the previous version of the results had. Let me know if this is possible or if there is a parameter to reveal this data on my results files. Thank you!

calebweinreb commented 11 months ago

Hi Ella, good question! We can put it back if there is demand for it. The reason I took it out is because the we don't validate in the paper that these are good estimates of the bodyparts per se and therefore didn't want to mislead people. Out of curiosity, what were you using them for and did you find them to be accurate?

At least having a flag to include them is a good idea. I can add that in the next small bugfix release this week!

ellaksay commented 11 months ago

We were using the nose keypoint to estimate distance from a port as another parameter to feed into an HMM alongside our keypoint_moseq clusters. Now knowing that the nose hasn't been validated, we can adjust our code so it uses the centroid instead because we didn't perform any cross-analysis to check how accurate the distances were. Thank you!