Update generated Python Op docs.

Change: 127274578

6 files changed, 85 insertions, 8 deletions

diff --git a/tensorflow/g3doc/api_docs/python/contrib.bayesflow.stochastic_graph.md b/tensorflow/g3doc/api_docs/python/contrib.bayesflow.stochastic_graph.md
index f8f0d6305c..27c46297c5 100644
--- a/tensorflow/g3doc/api_docs/python/contrib.bayesflow.stochastic_graph.md
+++ b/tensorflow/g3doc/api_docs/python/contrib.bayesflow.stochastic_graph.md
@@ -14,7 +14,7 @@ Classes and helper functions for Stochastic Computation Graphs.
 Base Class for Tensor-like objects that emit stochastic values.
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.StochasticTensor.__init__(**kwargs)` {#StochasticTensor.__init__}
+#### `tf.contrib.bayesflow.stochastic_graph.StochasticTensor.__init__()` {#StochasticTensor.__init__}
 
 
 
@@ -84,12 +84,37 @@ for purposes of the gradient.
 
 ### `class tf.contrib.bayesflow.stochastic_graph.DistributionTensor` {#DistributionTensor}
 
-The DistributionTensor is a StochasticTensor backed by a distribution.
+DistributionTensor is a StochasticTensor backed by a distribution.
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.__init__(dist_cls, name=None, dist_value_type=None, **dist_args)` {#DistributionTensor.__init__}
+#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.__init__(dist_cls, name=None, dist_value_type=None, surrogate_loss_fn=score_function, **dist_args)` {#DistributionTensor.__init__}
 
+Construct a `DistributionTensor`.
 
+`surrogate_loss_fn` controls what `surrogate_loss` returns, which is used
+in conjunction with the `surrogate_losses` function in this module.
+`surrogate_loss_fn` is a callable that takes this `DistributionTensor`, a
+`Tensor` with this `DistributionTensor`'s value, and a list of `Tensor`
+losses influenced by this `DistributionTensor`; it should return a `Tensor`
+surrogate loss. If not provided, it defaults to the score function
+surrogate loss: `log_prob(value) * sum(losses)`. If `surrogate_loss_fn` is
+None, no surrogate loss will be returned. Currently, a surrogate loss will
+only be used if `dist_value_type.stop_gradient=True` or if the value is a
+sample from a non-reparameterized distribution.
+
+##### Args:
+
+
+*  <b>`dist_cls`</b>: a class deriving from `BaseDistribution`.
+*  <b>`name`</b>: a name for this `DistributionTensor` and its ops.
+*  <b>`dist_value_type`</b>: a `_StochasticValueType`, which will determine what the
+      `value` of this `DistributionTensor` will be. If not provided, the
+      value type set with the `value_type` context manager will be used.
+*  <b>`surrogate_loss_fn`</b>: callable that takes
+      `(dt, dt.value(), influenced_losses)`, where `dt` is this
+      `DistributionTensor`, and returns a `Tensor` surrogate loss.
+*  <b>`**dist_args`</b>: keyword arguments to be passed through to `dist_cls` on
+      construction.
 
 
 - - -
@@ -150,7 +175,7 @@ The DistributionTensor is a StochasticTensor backed by a distribution.
 
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.surrogate_loss(losses, name=None)` {#DistributionTensor.surrogate_loss}
+#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.surrogate_loss(losses, name='DistributionSurrogateLoss')` {#DistributionTensor.surrogate_loss}
 
 
 
@@ -404,6 +429,23 @@ in a `stop_gradients` call to disable any possible backpropagation.
 
 
 
+## Stochastic Computation Surrogate Loss Functions
+
+- - -
+
+### `tf.contrib.bayesflow.stochastic_graph.score_function(dist_tensor, value, losses)` {#score_function}
+
+
+
+
+- - -
+
+### `tf.contrib.bayesflow.stochastic_graph.get_score_function_with_baseline(baseline)` {#get_score_function_with_baseline}
+
+
+
+
+
 ## Stochastic Computation Graph Helper Functions
 
 - - -
diff --git a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard0/tf.contrib.bayesflow.stochastic_graph.DistributionTensor.md b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard0/tf.contrib.bayesflow.stochastic_graph.DistributionTensor.md
index 3bb6771efa..8457b37301 100644
--- a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard0/tf.contrib.bayesflow.stochastic_graph.DistributionTensor.md
+++ b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard0/tf.contrib.bayesflow.stochastic_graph.DistributionTensor.md
@@ -1,9 +1,34 @@
-The DistributionTensor is a StochasticTensor backed by a distribution.
+DistributionTensor is a StochasticTensor backed by a distribution.
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.__init__(dist_cls, name=None, dist_value_type=None, **dist_args)` {#DistributionTensor.__init__}
+#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.__init__(dist_cls, name=None, dist_value_type=None, surrogate_loss_fn=score_function, **dist_args)` {#DistributionTensor.__init__}
 
+Construct a `DistributionTensor`.
 
+`surrogate_loss_fn` controls what `surrogate_loss` returns, which is used
+in conjunction with the `surrogate_losses` function in this module.
+`surrogate_loss_fn` is a callable that takes this `DistributionTensor`, a
+`Tensor` with this `DistributionTensor`'s value, and a list of `Tensor`
+losses influenced by this `DistributionTensor`; it should return a `Tensor`
+surrogate loss. If not provided, it defaults to the score function
+surrogate loss: `log_prob(value) * sum(losses)`. If `surrogate_loss_fn` is
+None, no surrogate loss will be returned. Currently, a surrogate loss will
+only be used if `dist_value_type.stop_gradient=True` or if the value is a
+sample from a non-reparameterized distribution.
+
+##### Args:
+
+
+*  <b>`dist_cls`</b>: a class deriving from `BaseDistribution`.
+*  <b>`name`</b>: a name for this `DistributionTensor` and its ops.
+*  <b>`dist_value_type`</b>: a `_StochasticValueType`, which will determine what the
+      `value` of this `DistributionTensor` will be. If not provided, the
+      value type set with the `value_type` context manager will be used.
+*  <b>`surrogate_loss_fn`</b>: callable that takes
+      `(dt, dt.value(), influenced_losses)`, where `dt` is this
+      `DistributionTensor`, and returns a `Tensor` surrogate loss.
+*  <b>`**dist_args`</b>: keyword arguments to be passed through to `dist_cls` on
+      construction.
 
 
 - - -
@@ -64,7 +89,7 @@ The DistributionTensor is a StochasticTensor backed by a distribution.
 
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.surrogate_loss(losses, name=None)` {#DistributionTensor.surrogate_loss}
+#### `tf.contrib.bayesflow.stochastic_graph.DistributionTensor.surrogate_loss(losses, name='DistributionSurrogateLoss')` {#DistributionTensor.surrogate_loss}
 
 
 
diff --git a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard1/tf.contrib.bayesflow.stochastic_graph.StochasticTensor.md b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard1/tf.contrib.bayesflow.stochastic_graph.StochasticTensor.md
index d65e7b17a4..1cc163ce38 100644
--- a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard1/tf.contrib.bayesflow.stochastic_graph.StochasticTensor.md
+++ b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard1/tf.contrib.bayesflow.stochastic_graph.StochasticTensor.md
@@ -1,7 +1,7 @@
 Base Class for Tensor-like objects that emit stochastic values.
 - - -
 
-#### `tf.contrib.bayesflow.stochastic_graph.StochasticTensor.__init__(**kwargs)` {#StochasticTensor.__init__}
+#### `tf.contrib.bayesflow.stochastic_graph.StochasticTensor.__init__()` {#StochasticTensor.__init__}
 
 
 
diff --git a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard6/tf.contrib.bayesflow.stochastic_graph.get_score_function_with_baseline.md b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard6/tf.contrib.bayesflow.stochastic_graph.get_score_function_with_baseline.md
new file mode 100644
index 0000000000..db24b0265f
--- /dev/null
+++ b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard6/tf.contrib.bayesflow.stochastic_graph.get_score_function_with_baseline.md
@@ -0,0 +1,4 @@
+### `tf.contrib.bayesflow.stochastic_graph.get_score_function_with_baseline(baseline)` {#get_score_function_with_baseline}
+
+
+
diff --git a/tensorflow/g3doc/api_docs/python/functions_and_classes/shard8/tf.contrib.bayesflow.stochastic_graph.score_function.md b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard8/tf.contrib.bayesflow.stochastic_graph.score_function.md
new file mode 100644
index 0000000000..85d27fa280
--- /dev/null
+++ b/tensorflow/g3doc/api_docs/python/functions_and_classes/shard8/tf.contrib.bayesflow.stochastic_graph.score_function.md
@@ -0,0 +1,4 @@
+### `tf.contrib.bayesflow.stochastic_graph.score_function(dist_tensor, value, losses)` {#score_function}
+
+
+
diff --git a/tensorflow/g3doc/api_docs/python/index.md b/tensorflow/g3doc/api_docs/python/index.md
index e1a9354216..151032f61e 100644
--- a/tensorflow/g3doc/api_docs/python/index.md
+++ b/tensorflow/g3doc/api_docs/python/index.md
@@ -573,10 +573,12 @@
 * **[BayesFlow Stochastic Graph (contrib)](../../api_docs/python/contrib.bayesflow.stochastic_graph.md)**:
   * [`DistributionTensor`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#DistributionTensor)
   * [`get_current_value_type`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#get_current_value_type)
+  * [`get_score_function_with_baseline`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#get_score_function_with_baseline)
   * [`MeanValue`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#MeanValue)
   * [`NoValueTypeSetError`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#NoValueTypeSetError)
   * [`SampleAndReshapeValue`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#SampleAndReshapeValue)
   * [`SampleValue`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#SampleValue)
+  * [`score_function`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#score_function)
   * [`StochasticTensor`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#StochasticTensor)
   * [`surrogate_losses`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#surrogate_losses)
   * [`value_type`](../../api_docs/python/contrib.bayesflow.stochastic_graph.md#value_type)