Fixed default value for emission_dist parameter of ParamsGGSSM in generalized_gaussian_ssm/models.py to correctly be a lambda function that takes in mean and cov and returns a tfp MultivariateNormalFullCovariance distribution object.
Updated _condition_on() in generalized_gaussian_ssm/inference.py to compute the marginal log likelihood with respect to the emission_dist parameter specified in ParamsGGSSM object.
Added a simple unit test in generalized_gaussian_ssm/models_test.py to verify that emissions generated via Poisson process lead to larger marginal log-likelihood under CMGF with Poisson emission_dist when compared to the previous default of MVN emission_dist.
Description
emission_dist
parameter ofParamsGGSSM
ingeneralized_gaussian_ssm/models.py
to correctly be a lambda function that takes inmean
andcov
and returns a tfpMultivariateNormalFullCovariance
distribution object._condition_on()
ingeneralized_gaussian_ssm/inference.py
to compute the marginal log likelihood with respect to theemission_dist
parameter specified inParamsGGSSM
object.generalized_gaussian_ssm/models_test.py
to verify that emissions generated via Poisson process lead to larger marginal log-likelihood under CMGF with Poissonemission_dist
when compared to the previous default of MVNemission_dist
.Issue
237