Model specifications for benchmarking dataset

[NeuTTH]

Thanks for the great work! I am hoping to replicate some of the results in the paper with regards to the benchmarking datasets. It is mentioned in the applying keypoint-MoSeq section of the paper that:

The one exception was $\kappa$ which we adjusted separately for each dataset to ensure a median state duration of 400ms.

I am wondering if $\kappa$ is different while training on the benchmarking datasets showed in Fig. 6. Thank you very much!

[calebweinreb]

Here is a table of parameters we used for each dataset. Let me know if you need any other information!

Dataset	Latent dim	AR only	Full model
Overhead 2D open field	4	1e6	7e4
Below-floor 2D open field	6	1e6	1e3
3D open field	6	1e6	2e4
IMU dataset	4	2.5e6	8e4
Dopamine photometry	4	2e6	3e5
Rat7M	8	1e9	7e4
CalMS21	6	1e6	3.30e4
DeepEthogram	10	1e9	2e6