This update refactors the modeling code into separate arhmm,slds, and keypoint_slds modules. Each module has separate files for initialization, resampling, and log likelihood computation, and each relies on the code for the model(s) it builds on.
The following small changes to the model initialization method should be noted, as they are breaking with respect to the keypoint-moseq code: (1) the initialization method now takes a data dictionary containing Y, mask, and (optionally) conf rather than separate keyword arguments for each of these values and (2) the latent_dimension keyword argument is now instead specified in the latent_dim field of the ar_hypparams. A small complementary commit has been made to keypoint-moseq to reflect these alterations.
This update refactors the modeling code into separate
arhmm
,slds
, andkeypoint_slds
modules. Each module has separate files for initialization, resampling, and log likelihood computation, and each relies on the code for the model(s) it builds on.The following small changes to the model initialization method should be noted, as they are breaking with respect to the
keypoint-moseq
code: (1) the initialization method now takes adata
dictionary containingY
,mask
, and (optionally)conf
rather than separate keyword arguments for each of these values and (2) thelatent_dimension
keyword argument is now instead specified in thelatent_dim
field of thear_hypparams
. A small complementary commit has been made tokeypoint-moseq
to reflect these alterations.