A library for generative and sampling-based modeling
GenSn provides a handful of useful utilities and modules to straight-forwardly construct and combine probabilistic modeling components. At the top level, GenSn can be thought of as an extension to torch.distributions, providing a way to define trainable distributions. Trainable distributions are much like a hybrid of PyTorch modules and PyTorch distribution objects. It lets you easily define a distribution, complete with its trainable parameters.
The library name GenSn (pronounced gen-sen with a hard G as in get) in Japanese (源泉) can be translated as source (such as source of a stream of water).