Implement the compute_output_shape method on DenseVariational.

tensorflow_probability/python/layers/dense_variational_v2.py

@@ -142,6 +142,24 @@ def call(self, inputs):
 
     return outputs
 
+  def compute_output_shape(self, input_shape):
+    """
+    Computes the output shape of the layer.
+    Args:
+      input_shape: `TensorShape` or `list` of `TensorShape`
+        (only last dim is used)
+    Returns:
+      The output shape.
+    Raises:
+        ValueError: If the innermost dimension of `input_shape` is not defined.
+    """
+    input_shape = tf.TensorShape(input_shape)
+    input_shape = input_shape.with_rank_at_least(2)
+    if input_shape[-1] is None:
+      raise ValueError(
+          'The innermost dimension of input_shape must be defined, but saw: %s' % (
+              input_shape,))
+    return input_shape[:-1].concatenate(self.units)  
 
 def _make_kl_divergence_penalty(
     use_exact_kl=False,

tensorflow_probability/python/layers/dense_variational_v2_test.py

@@ -76,10 +76,11 @@ def test_end_to_end(self):
     # Get dataset.
     y, x, x_tst = create_dataset()
 
-    # Build model.
+    layer = tfp.layers.DenseVariational(1, posterior_mean_field, prior_trainable)
+
     model = tf.keras.Sequential([
-        tfp.layers.DenseVariational(1, posterior_mean_field, prior_trainable),
-        tfp.layers.DistributionLambda(lambda t: tfd.Normal(loc=t, scale=1)),
+        layer,
+        tfp.layers.DistributionLambda(lambda t: tfd.Normal(loc=t, scale=1))
     ])
 
     # Do inference.
@@ -96,6 +97,10 @@ def test_end_to_end(self):
     self.assertContainsSubsequence(posterior.name, '/posterior/')
     self.assertContainsSubsequence(prior.name, '/prior/')
 
+    # Check the output_shape.
+    expected_output_shape = layer.compute_output_shape((None, x.shape[-1]))
+    self.assertAllEqual(expected_output_shape, (None, 1))
+
     # Profit.
     yhat = model(x_tst)
     self.assertIsInstance(yhat, tfd.Distribution)