Pascal Schetelat

Hi, Some colleagues and I are really interested in this and we are willing to contribute. I tried the pull requests and noticed breaking issues that need correcting and testing....

Ok, I have a fisrt implementation using rejection sampling in order to take into account evidence on non marginal nodes. I have resulted to use networkx topological sort to order...

Ok, so rejection sampling was easy. Gibbs sampling, on the other hand require a little bit more work to build the transition kernel of the markov chains. Nothing impossible, but...

I started with rejection sampling wich is easy to implement but quite slow as soon as evidences are specified on non root nodes, due to a large rate of rejection....

By the way, I should probably have not relied on that assumption, but is there a good reason why distribution.keys() and keys in the distribution.keymap are sometimes ordered the same,...

I just made the PR here : https://github.com/jmschrei/pomegranate/pull/778

Hi Mattjin, Have you been able to replicate it on the other exemples or is it specific to this data set? Could you provide me with the code that generated...