DPP has a non-trainable parameter eigenvectors (unless/until we add Expm+FillSkewSymmetric bijectors).

PiperOrigin-RevId: 343490438

Author: brianwa84

tensorflow_probability/python/distributions/distribution_properties_test.py

@@ -289,6 +289,7 @@ def testCanConstructAndSampleDistribution(self, data):
     non_trainable_tensor_params = (
         'atol',
         'rtol',
+        'eigenvectors',  # TODO(b/171872834): DeterminantalPointProcess
         'total_count',
         'num_samples',
         'df',  # Can't represent constraint that Wishart df > dimension.