Hi there!
Does anyone know how can I implement my own MCMC kernel. Specifically I want to sample from the posterior distribution of a BNN but I want to use my own sampler. In the documentation it says that the kernel has to be an instance of the TraceKernel class. However, I cannot find any information of the requirements such class must have in order to implement mine. Does anyone have any idea on how can I do this? I thought of using the code for HMC or NUTS as a guide, given that it should imitate the workflow of any of those, but I am still unsure about what exactly I can modify or not.
Thanks in advance!