Prefer + to tf.add_n in JointDistributionPinned.log_weight, because it broadcasts.

PiperOrigin-RevId: 378889317

Author: axch

tensorflow_probability/python/experimental/distributions/joint_distribution_pinned.py

@@ -557,7 +557,7 @@ def log_weight(self, *args, **kwargs):  # pylint: disable=g-doc-args
       log_weights: log-weight of the given point, i.e. the log pinned evidence.
     """
     pin_probs = self.unnormalized_log_prob_parts(*args, **kwargs).pinned
-    return tf.add_n(
+    return sum(  # Sum uses +, which broadcasts
         pin_probs.values() if isinstance(pin_probs, dict) else pin_probs)
 
   @docstring_util.expand_docstring(