fix bug when semip is passed None for likelihood (facebookresearch#347)

Summary:
Pull Request resolved: facebookresearch#347

Adds a check to make sure likelihood class is not None when constructing the likelihood of SemiP from a config

Reviewed By: mshvartsman

Differential Revision: D57624855

fbshipit-source-id: 514996c4f2c97c98d33171e811199ad47466a98e

aepsych/models/semi_p.py

@@ -280,8 +280,10 @@ def from_config(cls, config: Config) -> SemiParametricGPModel:
 
         if hasattr(likelihood_cls, "from_config"):
             likelihood = likelihood_cls.from_config(config)
-        else:
+        elif likelihood_cls is not None:
             likelihood = likelihood_cls()
+        else:
+            likelihood = None
 
         stim_dim = config.getint(classname, "stim_dim", fallback=0)