zhusuan
zhusuan copied to clipboard
FlowDistribution
Following the conversation in #70
Currently, I have only modified the planar_flow
to be single transformation and to return directly the log_det
of the transformation. I'm not too sure if the way I've used variable scopes is currently correct, so any suggestions on that would be nice. I need to also to change accordingly and the other flows. I've actually have implemented before quite a few so would be happy to add more.
Additionally, discussion on wether you guys like the FlowDistribution
abstraction is needed. Tests for that have to be added as well.
Cool, I will start looking at this once I finished the matrix normal one.
Back from rushing an icml paper. Sorry for being late but I'm reading the code now.
Sure no worries, hope ICML went well.