Update the release notes with information about tf.data.

Also adds a short porting guide to the tf.contrib.data README.

PiperOrigin-RevId: 171097798

RELEASE.md

@@ -1,6 +1,16 @@
 # Release 1.4.0
 
 ## Major Features And Improvements
+* `tf.data` is now part of the core TensorFlow API.
+  * The API is now subject to backwards compatibility guarantees.
+  * For a guide to migrating from the `tf.contrib.data` API, see the
+    [README] (https://github.com/tensorflow/tensorflow/blob/r1.4/tensorflow/contrib/data/README.md).
+  * Major new features include `Dataset.from_generator()` (for building an input
+    pipeline from a Python generator), and the `Dataset.apply()` method for
+    applying custom transformation functions.
+  * Several custom transformation functions have been added, including
+    `tf.contrib.data.batch_and_drop_remainder()` and
+    `tf.contrib.data.sloppy_interleave()`.
 * Java:
   * Generics (e.g., `Tensor<Integer>`) for improved type-safety (courtesy @andrewcmyers).
   * Support for multi-dimensional string tensors.
@@ -16,6 +26,11 @@
   flexible and reproducible package, is available via the new
   `tf.contrib.data.Dataset.from_generator` method!
 
+## Breaking Changes to the API
+* The signature of the `tf.contrib.data.rejection_resample()` function has been
+  changed. It now returns a function that can be used as an argument to
+  `Dataset.apply()`.
+
 # Release 1.3.0
 
 See also [TensorBoard 0.1.4](https://github.com/tensorflow/tensorboard/releases/tag/0.1.4) release notes.

tensorflow/contrib/data/README.md

@@ -2,9 +2,38 @@
 =====================
 
 NOTE: The `tf.contrib.data` module has been deprecated. Use `tf.data` instead.
+We are continuing to support existing code using the `tf.contrib.data` APIs in
+the current version of TensorFlow, but will eventually remove support. The
+`tf.data` APIs are subject to backwards compatibility guarantees.
 
-This directory contains the Python API for the `tf.contrib.data.Dataset` and
-`tf.contrib.data.Iterator` classes, which can be used to build input pipelines.
+Porting your code to `tf.data`
+------------------------------
 
-The documentation for `tf.data` API has moved to the programmers'
-guide, [here](../../docs_src/programmers_guide/datasets.md).
+The `tf.contrib.data.Dataset` class has been renamed to `tf.data.Dataset`, and
+the `tf.contrib.data.Iterator` class has been renamed to `tf.data.Iterator`.
+Most code can be ported by removing `.contrib` from the names of the classes.
+However, there are some small differences, which are outlined below.
+
+The arguments accepted by the `Dataset.map()` transformation have changed:
+
+* `dataset.map(..., num_threads=T)` is now `dataset.map(num_parallel_calls=T)`.
+* `dataset.map(..., output_buffer_size=B)` is now
+  `dataset.map(...).prefetch(B).
+
+Some transformations have been removed from `tf.data.Dataset`, and you must
+instead apply them using `Dataset.apply()` transformation. The full list of
+changes is as follows:
+
+* `dataset.dense_to_sparse_batch(...)` is now
+  `dataset.apply(tf.contrib.data.dense_to_sparse_batch(...)`.
+* `dataset.enumerate(...)` is now
+  `dataset.apply(tf.contrib.data.enumerate_dataset(...))`.
+* `dataset.group_by_window(...)` is now
+  `dataset.apply(tf.contrib.data.group_by_window(...))`.
+* `dataset.ignore_errors()` is now
+  `dataset.apply(tf.contrib.data.ignore_errors())`.
+* `dataset.unbatch()` is now `dataset.apply(tf.contrib.data.unbatch())`.
+
+The `Dataset.make_dataset_resource()` and `Iterator.dispose_op()` methods have
+been removed from the API. Please open a GitHub issue if you have a need for
+either of these.