search

YttriLab / B-SOID

Behavioral segmentation of open field in DeepLabCut, or B-SOID ("B-side"), is a pipeline that pairs unsupervised pattern recognition with supervised classification to achieve fast predictions of behaviors that are not predefined by users.
GNU General Public License v3.0
183 stars 51 forks source link

UMAP cluster coordinates as ouput #80

Open ellaksay opened 10 months ago

ellaksay commented 10 months ago

Hello! I was wondering if there was a way the output csv files could be modified to include the coordinates where the cluster would map onto the umap (for example, when mapping on the side-by-side video). There is a "sub-cluster" of one of my clusters that corresponds and consistently is mapped onto during a certain behavior and I think setting a region of interest around those coordinates would be helpful to quantify behavior rather than troubleshooting new clusters and hoping to get one that is classified in that area. Let me know if this is a possibility! Otherwise, do you have recommendations to making my clusters more specific if currently they appear to be too general? I can send more data if needed. Thank you.

39gofpnut commented 9 months ago 
 @ellaksay hi! I just started using bsoid and I have some questions regarding it. I think your knowledge and familiarity with the software would help me a lot! Do you have socials (Twitter, etc) that I can personally contact directly? Looking forward to your reply!
Kagiro-K commented 6 months ago

I also have the same question! Would appreciate an update, please! :)

IsabelleSajonia commented 3 months ago

I'm not sure I fully understand the question but are you just trying to quantify a particular cluster? I made a pie chart in R that calculates the percentage of time an animal spends performing a certain behavior if that would help. As far as UMAP coordinates go, I used the embeddings and cluster sav to make my own UMAP, you might be able to add a line to that to save coordinates (rather than working in the API). Since you've used this pipeline, can I ask if you've ever run this on multiple datafiles? When I run the API with multiple datafiles only one seems to ever end up in the sav file.

Sathyanesan-Lab commented 3 weeks ago

Hi @IsabelleSajonia we have been trying to do exactly this since the UMAP generated by the app always has the legend on top of the graph making it harder to see some clusters. Can you share how exactly you generate your own UMAPs using the sav file?