I have read docs and examples, but may miss a Tutorials about how to code custom distributions by myself. Would someone mind to send me a Tutorials that use transforms to code params and methods in my distributions. I want to create GMM distributions like as already created GSM class, and inference this model by SVI.
Thanks.
Hi @HerdingCat,
I would recommend forking-and-modifying one of the existing distributions in either torch.distributions (many of which inherit from TransformedDistribution) or pyro.distributions. Also feel free to add an issue requesting a distributions tutorial.