Allow fixing of arbitrary hyperparameters

[hadsed]

This may or may not be a valid thing to do, but sometimes when experimenting one would like to fix the hyperparameters from going nuts to see general behavior. This is especially true for a hyperparameter corresponding to a noise variable, like in the likelihood function.