Modify docs under contrib/distributions to point to tfp.

PiperOrigin-RevId: 213866466

33 files changed, 118 insertions, 74 deletions

diff --git a/tensorflow/contrib/distributions/python/ops/autoregressive.py b/tensorflow/contrib/distributions/python/ops/autoregressive.py
index bb9b8043b2..3ba1c3a665 100644
--- a/tensorflow/contrib/distributions/python/ops/autoregressive.py
+++ b/tensorflow/contrib/distributions/python/ops/autoregressive.py
@@ -65,13 +65,14 @@ class Autoregressive(distribution_lib.Distribution):
   ```
 
   where the ellipses (`...`) represent `n-2` composed calls to `fn`, `fn`
-  constructs a `tf.distributions.Distribution`-like instance, and `x0` is a
+  constructs a `tfp.distributions.Distribution`-like instance, and `x0` is a
   fixed initializing `Tensor`.
 
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   def normal_fn(self, event_size):
     n = event_size * (event_size + 1) / 2
@@ -127,7 +128,7 @@ class Autoregressive(distribution_lib.Distribution):
 
     Args:
       distribution_fn: Python `callable` which constructs a
-        `tf.distributions.Distribution`-like instance from a `Tensor` (e.g.,
+        `tfp.distributions.Distribution`-like instance from a `Tensor` (e.g.,
         `sample0`). The function must respect the "autoregressive property",
         i.e., there exists a permutation of event such that each coordinate is a
         diffeomorphic function of on preceding coordinates.
diff --git a/tensorflow/contrib/distributions/python/ops/batch_reshape.py b/tensorflow/contrib/distributions/python/ops/batch_reshape.py
index 519077bc9a..612376efb7 100644
--- a/tensorflow/contrib/distributions/python/ops/batch_reshape.py
+++ b/tensorflow/contrib/distributions/python/ops/batch_reshape.py
@@ -45,7 +45,8 @@ class BatchReshape(distribution_lib.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   dtype = np.float32
   dims = 2
diff --git a/tensorflow/contrib/distributions/python/ops/bijectors/masked_autoregressive.py b/tensorflow/contrib/distributions/python/ops/bijectors/masked_autoregressive.py
index 296e66f2b2..3b3d8ee6f2 100644
--- a/tensorflow/contrib/distributions/python/ops/bijectors/masked_autoregressive.py
+++ b/tensorflow/contrib/distributions/python/ops/bijectors/masked_autoregressive.py
@@ -61,8 +61,8 @@ class MaskedAutoregressiveFlow(bijector.Bijector):
   `shift_and_log_scale_fn`, `masked_autoregressive_default_template`, achieves
   this property by zeroing out weights in its `masked_dense` layers.
 
-  In the `tf.distributions` framework, a "normalizing flow" is implemented as a
-  `tf.contrib.distributions.bijectors.Bijector`. The `forward` "autoregression"
+  In the `tfp` framework, a "normalizing flow" is implemented as a
+  `tfp.bijectors.Bijector`. The `forward` "autoregression"
   is implemented using a `tf.while_loop` and a deep neural network (DNN) with
   masked weights such that the autoregressive property is automatically met in
   the `inverse`.
@@ -126,8 +126,9 @@ class MaskedAutoregressiveFlow(bijector.Bijector):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
-  tfb = tfd.bijectors
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+  tfb = tfp.bijectors
 
   dims = 5
 
diff --git a/tensorflow/contrib/distributions/python/ops/bijectors/permute.py b/tensorflow/contrib/distributions/python/ops/bijectors/permute.py
index f182a1adcb..178c3c94bf 100644
--- a/tensorflow/contrib/distributions/python/ops/bijectors/permute.py
+++ b/tensorflow/contrib/distributions/python/ops/bijectors/permute.py
@@ -41,9 +41,10 @@ class Permute(bijector.Bijector):
   """Permutes the rightmost dimension of a `Tensor`.
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfb = tfp.bijectors
 
-  reverse = tfd.bijectors.Permute(permutation=[2, 1, 0])
+  reverse = tfb.Permute(permutation=[2, 1, 0])
 
   reverse.forward([-1., 0., 1.])
   # ==> [1., 0., -1]
diff --git a/tensorflow/contrib/distributions/python/ops/bijectors/real_nvp.py b/tensorflow/contrib/distributions/python/ops/bijectors/real_nvp.py
index 773ae24461..0bcb08cdea 100644
--- a/tensorflow/contrib/distributions/python/ops/bijectors/real_nvp.py
+++ b/tensorflow/contrib/distributions/python/ops/bijectors/real_nvp.py
@@ -90,8 +90,9 @@ class RealNVP(bijector.Bijector):
   #### Example Use
 
   ```python
-  tfd = tf.contrib.distributions
-  tfb = tfd.bijectors
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+  tfb = tfp.bijectors
 
   # A common choice for a normalizing flow is to use a Gaussian for the base
   # distribution. (However, any continuous distribution would work.) E.g.,
diff --git a/tensorflow/contrib/distributions/python/ops/bijectors/reshape.py b/tensorflow/contrib/distributions/python/ops/bijectors/reshape.py
index c8282229a3..71ac29038f 100644
--- a/tensorflow/contrib/distributions/python/ops/bijectors/reshape.py
+++ b/tensorflow/contrib/distributions/python/ops/bijectors/reshape.py
@@ -80,9 +80,10 @@ class Reshape(bijector.Bijector):
   Example usage:
   ```python
 
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfb = tfp.bijectors
 
-  r = tfd.bijectors.Reshape(event_shape_out=[1, -1])
+  r = tfb.Reshape(event_shape_out=[1, -1])
 
   r.forward([3., 4.])    # shape [2]
   # ==> [[3., 4.]]       # shape [1, 2]
diff --git a/tensorflow/contrib/distributions/python/ops/bijectors/scale_tril.py b/tensorflow/contrib/distributions/python/ops/bijectors/scale_tril.py
index 6fbe866578..0a6d690b65 100644
--- a/tensorflow/contrib/distributions/python/ops/bijectors/scale_tril.py
+++ b/tensorflow/contrib/distributions/python/ops/bijectors/scale_tril.py
@@ -42,7 +42,10 @@ class ScaleTriL(chain.Chain):
   #### Examples
 
   ```python
-  tfb = tf.contrib.distributions.bijectors
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+  tfb = tfp.bijectors
+
   b = tfb.ScaleTriL(
        diag_bijector=tfb.Exp(),
        diag_shift=None)
diff --git a/tensorflow/contrib/distributions/python/ops/cauchy.py b/tensorflow/contrib/distributions/python/ops/cauchy.py
index cb5223b055..c461833b9a 100644
--- a/tensorflow/contrib/distributions/python/ops/cauchy.py
+++ b/tensorflow/contrib/distributions/python/ops/cauchy.py
@@ -63,7 +63,8 @@ class Cauchy(distribution.Distribution):
   Examples of initialization of one or a batch of distributions.
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Define a single scalar Cauchy distribution.
   dist = tfd.Cauchy(loc=0., scale=3.)
diff --git a/tensorflow/contrib/distributions/python/ops/deterministic.py b/tensorflow/contrib/distributions/python/ops/deterministic.py
index affc64a14f..507c5d3679 100644
--- a/tensorflow/contrib/distributions/python/ops/deterministic.py
+++ b/tensorflow/contrib/distributions/python/ops/deterministic.py
@@ -198,8 +198,11 @@ class Deterministic(_BaseDeterministic):
   #### Examples
 
   ```python
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   # Initialize a single Deterministic supported at zero.
-  constant = tf.contrib.distributions.Deterministic(0.)
+  constant = tfd.Deterministic(0.)
   constant.prob(0.)
   ==> 1.
   constant.prob(2.)
@@ -208,7 +211,7 @@ class Deterministic(_BaseDeterministic):
   # Initialize a [2, 2] batch of scalar constants.
   loc = [[0., 1.], [2., 3.]]
   x = [[0., 1.1], [1.99, 3.]]
-  constant = tf.contrib.distributions.Deterministic(loc)
+  constant = tfd.Deterministic(loc)
   constant.prob(x)
   ==> [[1., 0.], [0., 1.]]
   ```
@@ -310,7 +313,8 @@ class VectorDeterministic(_BaseDeterministic):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single VectorDeterministic supported at [0., 2.] in R^2.
   constant = tfd.Deterministic([0., 2.])
diff --git a/tensorflow/contrib/distributions/python/ops/gumbel.py b/tensorflow/contrib/distributions/python/ops/gumbel.py
index acdea4d61d..4b50df5b48 100644
--- a/tensorflow/contrib/distributions/python/ops/gumbel.py
+++ b/tensorflow/contrib/distributions/python/ops/gumbel.py
@@ -63,7 +63,8 @@ class _Gumbel(distribution.Distribution):
   Examples of initialization of one or a batch of distributions.
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Define a single scalar Gumbel distribution.
   dist = tfd.Gumbel(loc=0., scale=3.)
diff --git a/tensorflow/contrib/distributions/python/ops/half_normal.py b/tensorflow/contrib/distributions/python/ops/half_normal.py
index b02c403106..f121637086 100644
--- a/tensorflow/contrib/distributions/python/ops/half_normal.py
+++ b/tensorflow/contrib/distributions/python/ops/half_normal.py
@@ -66,15 +66,18 @@ class HalfNormal(distribution.Distribution):
   Examples of initialization of one or a batch of distributions.
 
   ```python
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   # Define a single scalar HalfNormal distribution.
-  dist = tf.contrib.distributions.HalfNormal(scale=3.0)
+  dist = tfd.HalfNormal(scale=3.0)
 
   # Evaluate the cdf at 1, returning a scalar.
   dist.cdf(1.)
 
   # Define a batch of two scalar valued HalfNormals.
   # The first has scale 11.0, the second 22.0
-  dist = tf.contrib.distributions.HalfNormal(scale=[11.0, 22.0])
+  dist = tfd.HalfNormal(scale=[11.0, 22.0])
 
   # Evaluate the pdf of the first distribution on 1.0, and the second on 1.5,
   # returning a length two tensor.
diff --git a/tensorflow/contrib/distributions/python/ops/independent.py b/tensorflow/contrib/distributions/python/ops/independent.py
index 0672702b96..e1cfff3c66 100644
--- a/tensorflow/contrib/distributions/python/ops/independent.py
+++ b/tensorflow/contrib/distributions/python/ops/independent.py
@@ -70,7 +70,8 @@ class Independent(distribution_lib.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Make independent distribution from a 2-batch Normal.
   ind = tfd.Independent(
diff --git a/tensorflow/contrib/distributions/python/ops/inverse_gamma.py b/tensorflow/contrib/distributions/python/ops/inverse_gamma.py
index 70d050d7a6..452628257e 100644
--- a/tensorflow/contrib/distributions/python/ops/inverse_gamma.py
+++ b/tensorflow/contrib/distributions/python/ops/inverse_gamma.py
@@ -89,7 +89,9 @@ class InverseGamma(distribution.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   dist = tfd.InverseGamma(concentration=3.0, rate=2.0)
   dist2 = tfd.InverseGamma(concentration=[3.0, 4.0], rate=[2.0, 3.0])
   ```
diff --git a/tensorflow/contrib/distributions/python/ops/logistic.py b/tensorflow/contrib/distributions/python/ops/logistic.py
index 02e3bad51e..21c9b5a354 100644
--- a/tensorflow/contrib/distributions/python/ops/logistic.py
+++ b/tensorflow/contrib/distributions/python/ops/logistic.py
@@ -61,7 +61,8 @@ class Logistic(distribution.Distribution):
   Examples of initialization of one or a batch of distributions.
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Define a single scalar Logistic distribution.
   dist = tfd.Logistic(loc=0., scale=3.)
diff --git a/tensorflow/contrib/distributions/python/ops/mixture.py b/tensorflow/contrib/distributions/python/ops/mixture.py
index 3b7114ef06..52b67f2c54 100644
--- a/tensorflow/contrib/distributions/python/ops/mixture.py
+++ b/tensorflow/contrib/distributions/python/ops/mixture.py
@@ -50,7 +50,9 @@ class Mixture(distribution.Distribution):
 
   ```python
   # Create a mixture of two Gaussians:
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   mix = 0.3
   bimix_gauss = tfd.Mixture(
     cat=tfd.Categorical(probs=[mix, 1.-mix]),
diff --git a/tensorflow/contrib/distributions/python/ops/mixture_same_family.py b/tensorflow/contrib/distributions/python/ops/mixture_same_family.py
index 8ffee940d0..f4d394ff29 100644
--- a/tensorflow/contrib/distributions/python/ops/mixture_same_family.py
+++ b/tensorflow/contrib/distributions/python/ops/mixture_same_family.py
@@ -44,7 +44,8 @@ class MixtureSameFamily(distribution.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   ### Create a mixture of two scalar Gaussians:
 
@@ -113,12 +114,12 @@ class MixtureSameFamily(distribution.Distribution):
     """Construct a `MixtureSameFamily` distribution.
 
     Args:
-      mixture_distribution: `tf.distributions.Categorical`-like instance.
+      mixture_distribution: `tfp.distributions.Categorical`-like instance.
         Manages the probability of selecting components. The number of
         categories must match the rightmost batch dimension of the
         `components_distribution`. Must have either scalar `batch_shape` or
         `batch_shape` matching `components_distribution.batch_shape[:-1]`.
-      components_distribution: `tf.distributions.Distribution`-like instance.
+      components_distribution: `tfp.distributions.Distribution`-like instance.
         Right-most batch dimension indexes components.
       validate_args: Python `bool`, default `False`. When `True` distribution
         parameters are checked for validity despite possibly degrading runtime
diff --git a/tensorflow/contrib/distributions/python/ops/mvn_diag.py b/tensorflow/contrib/distributions/python/ops/mvn_diag.py
index cd0c282ba6..0b5b76be92 100644
--- a/tensorflow/contrib/distributions/python/ops/mvn_diag.py
+++ b/tensorflow/contrib/distributions/python/ops/mvn_diag.py
@@ -85,7 +85,8 @@ class MultivariateNormalDiag(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 2-variate Gaussian.
   mvn = tfd.MultivariateNormalDiag(
diff --git a/tensorflow/contrib/distributions/python/ops/mvn_diag_plus_low_rank.py b/tensorflow/contrib/distributions/python/ops/mvn_diag_plus_low_rank.py
index 74d9d04fc7..80546083d3 100644
--- a/tensorflow/contrib/distributions/python/ops/mvn_diag_plus_low_rank.py
+++ b/tensorflow/contrib/distributions/python/ops/mvn_diag_plus_low_rank.py
@@ -87,7 +87,8 @@ class MultivariateNormalDiagPlusLowRank(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate Gaussian with covariance `cov = S @ S.T`,
   # `S = diag(d) + U @ diag(m) @ U.T`. The perturbation, `U @ diag(m) @ U.T`, is
diff --git a/tensorflow/contrib/distributions/python/ops/mvn_full_covariance.py b/tensorflow/contrib/distributions/python/ops/mvn_full_covariance.py
index dbc4c1b3dc..bcb4937980 100644
--- a/tensorflow/contrib/distributions/python/ops/mvn_full_covariance.py
+++ b/tensorflow/contrib/distributions/python/ops/mvn_full_covariance.py
@@ -73,7 +73,8 @@ class MultivariateNormalFullCovariance(mvn_tril.MultivariateNormalTriL):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate Gaussian.
   mu = [1., 2, 3]
diff --git a/tensorflow/contrib/distributions/python/ops/mvn_linear_operator.py b/tensorflow/contrib/distributions/python/ops/mvn_linear_operator.py
index efe5a6d0d9..8fdc99824b 100644
--- a/tensorflow/contrib/distributions/python/ops/mvn_linear_operator.py
+++ b/tensorflow/contrib/distributions/python/ops/mvn_linear_operator.py
@@ -91,7 +91,8 @@ class MultivariateNormalLinearOperator(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate Gaussian.
   mu = [1., 2, 3]
diff --git a/tensorflow/contrib/distributions/python/ops/mvn_tril.py b/tensorflow/contrib/distributions/python/ops/mvn_tril.py
index c6a23e4336..c21f70fc3b 100644
--- a/tensorflow/contrib/distributions/python/ops/mvn_tril.py
+++ b/tensorflow/contrib/distributions/python/ops/mvn_tril.py
@@ -77,13 +77,14 @@ class MultivariateNormalTriL(
   ```
 
   Trainable (batch) lower-triangular matrices can be created with
-  `tf.contrib.distributions.matrix_diag_transform()` and/or
-  `tf.contrib.distributions.fill_triangular()`
+  `tfp.distributions.matrix_diag_transform()` and/or
+  `tfp.distributions.fill_triangular()`
 
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate Gaussian.
   mu = [1., 2, 3]
diff --git a/tensorflow/contrib/distributions/python/ops/poisson_lognormal.py b/tensorflow/contrib/distributions/python/ops/poisson_lognormal.py
index 7a7ad1be35..85683e3233 100644
--- a/tensorflow/contrib/distributions/python/ops/poisson_lognormal.py
+++ b/tensorflow/contrib/distributions/python/ops/poisson_lognormal.py
@@ -220,7 +220,8 @@ class PoissonLogNormalQuadratureCompound(distribution_lib.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Create two batches of PoissonLogNormalQuadratureCompounds, one with
   # prior `loc = 0.` and another with `loc = 1.` In both cases `scale = 1.`
diff --git a/tensorflow/contrib/distributions/python/ops/quantized_distribution.py b/tensorflow/contrib/distributions/python/ops/quantized_distribution.py
index 18a0f754e6..134658deab 100644
--- a/tensorflow/contrib/distributions/python/ops/quantized_distribution.py
+++ b/tensorflow/contrib/distributions/python/ops/quantized_distribution.py
@@ -196,8 +196,9 @@ class QuantizedDistribution(distributions.Distribution):
   parameter determining the unnormalized probability of that component.
 
   ```python
-  tfd = tf.contrib.distributions
-  tfb = tfd.bijectors
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+  tfb = tfp.bijectors
 
   net = wavenet(inputs)
   loc, unconstrained_scale, logits = tf.split(net,
diff --git a/tensorflow/contrib/distributions/python/ops/sinh_arcsinh.py b/tensorflow/contrib/distributions/python/ops/sinh_arcsinh.py
index a9d0fb4ccf..4b520b912e 100644
--- a/tensorflow/contrib/distributions/python/ops/sinh_arcsinh.py
+++ b/tensorflow/contrib/distributions/python/ops/sinh_arcsinh.py
@@ -124,7 +124,7 @@ class SinhArcsinh(transformed_distribution.TransformedDistribution):
       tailweight:  Tailweight parameter. Default is `1.0` (unchanged tailweight)
       distribution: `tf.Distribution`-like instance. Distribution that is
         transformed to produce this distribution.
-        Default is `tf.distributions.Normal(0., 1.)`.
+        Default is `tfp.distributions.Normal(0., 1.)`.
         Must be a scalar-batch, scalar-event distribution.  Typically
         `distribution.reparameterization_type = FULLY_REPARAMETERIZED` or it is
         a function of non-trainable parameters. WARNING: If you backprop through
diff --git a/tensorflow/contrib/distributions/python/ops/statistical_testing.py b/tensorflow/contrib/distributions/python/ops/statistical_testing.py
index c25e8c51d7..af22f4843a 100644
--- a/tensorflow/contrib/distributions/python/ops/statistical_testing.py
+++ b/tensorflow/contrib/distributions/python/ops/statistical_testing.py
@@ -30,27 +30,27 @@ is some expected constant.  Suppose the support of P is the interval
 `[0, 1]`.  Then you might do this:
 
 ```python
-tfd = tf.contrib.distributions
-
-expected_mean = ...
-num_samples = 5000
-samples = ... draw 5000 samples from P
-
-# Check that the mean looks right
-check1 = tfd.assert_true_mean_equal_by_dkwm(
-    samples, low=0., high=1., expected=expected_mean,
-    false_fail_rate=1e-6)
-
-# Check that the difference in means detectable with 5000 samples is
-# small enough
-check2 = tf.assert_less(
-    tfd.min_discrepancy_of_true_means_detectable_by_dkwm(
-        num_samples, low=0., high=1.0,
-        false_fail_rate=1e-6, false_pass_rate=1e-6),
-    0.01)
-
-# Be sure to execute both assertion ops
-sess.run([check1, check2])
+  from tensorflow_probability.python.distributions.internal import statistical_testing
+
+  expected_mean = ...
+  num_samples = 5000
+  samples = ... draw 5000 samples from P
+
+  # Check that the mean looks right
+  check1 = statistical_testing.assert_true_mean_equal_by_dkwm(
+      samples, low=0., high=1., expected=expected_mean,
+      false_fail_rate=1e-6)
+
+  # Check that the difference in means detectable with 5000 samples is
+  # small enough
+  check2 = tf.assert_less(
+      statistical_testing.min_discrepancy_of_true_means_detectable_by_dkwm(
+          num_samples, low=0., high=1.0,
+          false_fail_rate=1e-6, false_pass_rate=1e-6),
+      0.01)
+
+  # Be sure to execute both assertion ops
+  sess.run([check1, check2])
 ```
 
 The second assertion is an instance of experiment design.  It's a
diff --git a/tensorflow/contrib/distributions/python/ops/vector_diffeomixture.py b/tensorflow/contrib/distributions/python/ops/vector_diffeomixture.py
index 3c8aae2797..a3d178357b 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_diffeomixture.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_diffeomixture.py
@@ -300,7 +300,8 @@ class VectorDiffeomixture(distribution_lib.Distribution):
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Create two batches of VectorDiffeomixtures, one with mix_loc=[0.],
   # another with mix_loc=[1]. In both cases, `K=2` and the affine
diff --git a/tensorflow/contrib/distributions/python/ops/vector_exponential_diag.py b/tensorflow/contrib/distributions/python/ops/vector_exponential_diag.py
index 73356a3625..36cbd71f8b 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_exponential_diag.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_exponential_diag.py
@@ -90,7 +90,8 @@ class VectorExponentialDiag(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 2-variate VectorExponential, supported on
   # {(x, y) in R^2 : x > 0, y > 0}.
diff --git a/tensorflow/contrib/distributions/python/ops/vector_exponential_linear_operator.py b/tensorflow/contrib/distributions/python/ops/vector_exponential_linear_operator.py
index 9a47b48557..fd5bf9ecc7 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_exponential_linear_operator.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_exponential_linear_operator.py
@@ -108,7 +108,8 @@ class VectorExponentialLinearOperator(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 2-variate VectorExponential, supported on
   # {(x, y) in R^2 : x > 0, y > 0}.
diff --git a/tensorflow/contrib/distributions/python/ops/vector_laplace_diag.py b/tensorflow/contrib/distributions/python/ops/vector_laplace_diag.py
index e68ddc569c..8cd4e128c7 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_laplace_diag.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_laplace_diag.py
@@ -102,7 +102,8 @@ class VectorLaplaceDiag(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 2-variate VectorLaplace.
   vla = tfd.VectorLaplaceDiag(
diff --git a/tensorflow/contrib/distributions/python/ops/vector_laplace_linear_operator.py b/tensorflow/contrib/distributions/python/ops/vector_laplace_linear_operator.py
index 3923161a33..67d2ccd28d 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_laplace_linear_operator.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_laplace_linear_operator.py
@@ -110,7 +110,8 @@ class VectorLaplaceLinearOperator(
   #### Examples
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate VectorLaplace with some desired covariance.
   mu = [1., 2, 3]
diff --git a/tensorflow/contrib/distributions/python/ops/vector_sinh_arcsinh_diag.py b/tensorflow/contrib/distributions/python/ops/vector_sinh_arcsinh_diag.py
index 49ffff24ca..da57d0cb55 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_sinh_arcsinh_diag.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_sinh_arcsinh_diag.py
@@ -152,7 +152,7 @@ class VectorSinhArcsinhDiag(transformed_distribution.TransformedDistribution):
         broadcastable with `event_shape`.
       distribution: `tf.Distribution`-like instance. Distribution from which `k`
         iid samples are used as input to transformation `F`.  Default is
-        `tf.distributions.Normal(loc=0., scale=1.)`.
+        `tfp.distributions.Normal(loc=0., scale=1.)`.
         Must be a scalar-batch, scalar-event distribution.  Typically
         `distribution.reparameterization_type = FULLY_REPARAMETERIZED` or it is
         a function of non-trainable parameters. WARNING: If you backprop through
diff --git a/tensorflow/contrib/distributions/python/ops/vector_student_t.py b/tensorflow/contrib/distributions/python/ops/vector_student_t.py
index f289b39e51..bad91a0844 100644
--- a/tensorflow/contrib/distributions/python/ops/vector_student_t.py
+++ b/tensorflow/contrib/distributions/python/ops/vector_student_t.py
@@ -92,7 +92,8 @@ class _VectorStudentT(transformed_distribution.TransformedDistribution):
   Extra leading dimensions, if provided, allow for batches.
 
   ```python
-  tfd = tf.contrib.distributions
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
 
   # Initialize a single 3-variate vector Student's t-distribution.
   mu = [1., 2, 3]
diff --git a/tensorflow/contrib/distributions/python/ops/wishart.py b/tensorflow/contrib/distributions/python/ops/wishart.py
index 49b9de0ab5..ee2fc58864 100644
--- a/tensorflow/contrib/distributions/python/ops/wishart.py
+++ b/tensorflow/contrib/distributions/python/ops/wishart.py
@@ -480,11 +480,14 @@ class WishartCholesky(_WishartLinearOperator):
   #### Examples
 
   ```python
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   # Initialize a single 3x3 Wishart with Cholesky factored scale matrix and 5
   # degrees-of-freedom.(*)
   df = 5
   chol_scale = tf.cholesky(...)  # Shape is [3, 3].
-  dist = tf.contrib.distributions.WishartCholesky(df=df, scale=chol_scale)
+  dist = tfd.WishartCholesky(df=df, scale=chol_scale)
 
   # Evaluate this on an observation in R^3, returning a scalar.
   x = ...  # A 3x3 positive definite matrix.
@@ -498,14 +501,14 @@ class WishartCholesky(_WishartLinearOperator):
   # Initialize two 3x3 Wisharts with Cholesky factored scale matrices.
   df = [5, 4]
   chol_scale = tf.cholesky(...)  # Shape is [2, 3, 3].
-  dist = tf.contrib.distributions.WishartCholesky(df=df, scale=chol_scale)
+  dist = tfd.WishartCholesky(df=df, scale=chol_scale)
 
   # Evaluate this on four observations.
   x = [[x0, x1], [x2, x3]]  # Shape is [2, 2, 3, 3].
   dist.prob(x)  # Shape is [2, 2].
 
   # (*) - To efficiently create a trainable covariance matrix, see the example
-  #   in tf.contrib.distributions.matrix_diag_transform.
+  #   in tfp.distributions.matrix_diag_transform.
   ```
 
   """
@@ -604,11 +607,14 @@ class WishartFull(_WishartLinearOperator):
   #### Examples
 
   ```python
+  import tensorflow_probability as tfp
+  tfd = tfp.distributions
+
   # Initialize a single 3x3 Wishart with Full factored scale matrix and 5
   # degrees-of-freedom.(*)
   df = 5
   scale = ...  # Shape is [3, 3]; positive definite.
-  dist = tf.contrib.distributions.WishartFull(df=df, scale=scale)
+  dist = tfd.WishartFull(df=df, scale=scale)
 
   # Evaluate this on an observation in R^3, returning a scalar.
   x = ...  # A 3x3 positive definite matrix.
@@ -622,14 +628,14 @@ class WishartFull(_WishartLinearOperator):
   # Initialize two 3x3 Wisharts with Full factored scale matrices.
   df = [5, 4]
   scale = ...  # Shape is [2, 3, 3].
-  dist = tf.contrib.distributions.WishartFull(df=df, scale=scale)
+  dist = tfd.WishartFull(df=df, scale=scale)
 
   # Evaluate this on four observations.
   x = [[x0, x1], [x2, x3]]  # Shape is [2, 2, 3, 3]; xi is positive definite.
   dist.prob(x)  # Shape is [2, 2].
 
   # (*) - To efficiently create a trainable covariance matrix, see the example
-  #   in tf.contrib.distributions.matrix_diag_transform.
+  #   in tfd.matrix_diag_transform.
   ```
 
   """