Refactors inference code for SLDS and keypoint-SLDS, allowing it to interface with the updated LGSSM inference code of dynamax, in particular to utilize the new parallelized Kalman filter implementation.
Adds a parallel_kalman flag to resample_model functions and kalman_sample that uses the parallel posterior sampler of probml/dynamax#324.
Adds a jitter parameter to stabilize sampling of the covariance matrix as in probml/dynamax#321.
Updated kalman_sample parameterization and docs.
Allowed exclusion of outlier frames during PCA fit.
Refactors inference code for SLDS and keypoint-SLDS, allowing it to interface with the updated LGSSM inference code of dynamax, in particular to utilize the new parallelized Kalman filter implementation.
parallel_kalmanflag toresample_modelfunctions andkalman_samplethat uses the parallel posterior sampler of probml/dynamax#324.jitterparameter to stabilize sampling of the covariance matrix as in probml/dynamax#321.kalman_sampleparameterization and docs.Also see #7 for more details.
TODO: