Fix nan gradients in TraceEnum_ELBO when evaluated eagerly (#544)

ordabayevy · web-flow · commit cbbf888d6a7a · 2021-08-03T11:01:18.000-04:00
* fix eager adjoint

* fix MarkovProduct

* revert sum_product

* revert xfail test
diff --git a/funsor/adjoint.py b/funsor/adjoint.py
@@ -58,7 +58,8 @@ def interpret(self, cls, *args):
             )
             for arg in args
         ]
-        self._eager_to_lazy[result] = reflect.interpret(cls, *lazy_args)
+        with self._old_interpretation:
+            self._eager_to_lazy[result] = reflect.interpret(cls, *lazy_args)
         return result
 
     def __enter__(self):

Original file line number	Diff line number	Diff line change
`@@ -58,7 +58,8 @@ def interpret(self, cls, *args):`
`58`	`58`	`)`
`59`	`59`	`for arg in args`
`60`	`60`	`]`
`61`		`- self._eager_to_lazy[result] = reflect.interpret(cls, *lazy_args)`
	`61`	`+ with self._old_interpretation:`
	`62`	`+ self._eager_to_lazy[result] = reflect.interpret(cls, *lazy_args)`
`62`	`63`	`return result`
`63`	`64`
`64`	`65`	`def __enter__(self):`