89 fixes the inference tests, but the log option for hierarchical_likelihood breaks the gradients whenever PE samples or injections lie outside of the population's support. Since we haven't seen any overflow/underflow issues with the non-loged likelihood, we'll stick with it for now.
89 fixes the inference tests, but the
logoption forhierarchical_likelihoodbreaks the gradients whenever PE samples or injections lie outside of the population's support. Since we haven't seen any overflow/underflow issues with the non-loged likelihood, we'll stick with it for now.