I’m using SVI to optimize a model. I’m slightly confused about the “stable_update” argument. When set to “True”, I still see that the loss is “nan”. The documentation says the returned state will return the current state, and not one that yielded “nan” or invalid values, but is this reflected in the loss that is printed out in the progress bar?
After looking into this more, the loss still prints out as “nan”, but the parameters outputted remain valid.