# Module: tf.contrib.deprecated

<br />

|---|
| [![](https://www.tensorflow.org/images/GitHub-Mark-32px.png) View source on GitHub](https://github.com/tensorflow/tensorflow/blob/v1.15.0/tensorflow/contrib/deprecated/__init__.py) |

Non-core alias for the deprecated tf.X_summary ops.

For TensorFlow 1.0, we have reorganized the TensorFlow summary ops into a
submodule, and made some semantic tweaks. The first thing to note is that we
moved the APIs around as follows:

    tf.scalar_summary -> tf.compat.v1.summary.scalar
    tf.histogram_summary -> tf.compat.v1.summary.histogram
    tf.audio_summary -> tf.compat.v1.summary.audio
    tf.image_summary -> tf.compat.v1.summary.image
    tf.merge_summary -> tf.compat.v1.summary.merge
    tf.merge_all_summaries -> tf.compat.v1.summary.merge_all

We think this API is cleaner and will improve long-term discoverability and
clarity of the TensorFlow API. But we also took the opportunity to make an
important change to how summary "tags" work. The "tag" of a summary is the
string that is associated with the output data, i.e. the key for organizing the
generated protobufs.

Previously, the tag was allowed to be any unique string; it had no relation
to the summary op generating it, and no relation to the TensorFlow name system.
This behavior made it very difficult to write reusable that would add
summary ops to the graph. If you had a function to add summary ops, you would
need to pass in a [`tf.name_scope`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/name_scope), manually, to that function to create
deduplicated tags. Otherwise your program would fail with a runtime error due
to tag collision.

The new summary APIs under [`tf.summary`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/summary) throw away the "tag" as an independent
concept; instead, the first argument is the node name. So summary tags now
automatically inherit the surrounding [`tf.name_scope`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/name_scope), and automatically
are deduplicated if there is a conflict. Now however, the only allowed
characters are alphanumerics, underscores, and forward slashes. To make
migration easier, the new APIs automatically convert illegal characters to
underscores.

Just as an example, consider the following "before" and "after" code snippets:

    # Before
    def add_activation_summaries(v, scope):
      tf.scalar_summary("%s/fraction_of_zero" % scope, tf.nn.fraction_of_zero(v))
      tf.histogram_summary("%s/activations" % scope, v)

    # After
    def add_activation_summaries(v):
      tf.compat.v1.summary.scalar("fraction_of_zero", tf.nn.fraction_of_zero(v))
      tf.compat.v1.summary.histogram("activations", v)

Now, so long as the add_activation_summaries function is called from within the
right [`tf.name_scope`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/name_scope), the behavior is the same.

Because this change does modify the behavior and could break tests, we can't
automatically migrate usage to the new APIs. That is why we are making the old
APIs temporarily available here at [`tf.contrib.deprecated`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated).

In addition to the name change described above, there are two further changes
to the new summary ops:

- the "max_images" argument for `tf.image_summary` was renamed to "max_outputs for [`tf.compat.v1.summary.image`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/summary/image)
- `tf.scalar_summary` accepted arbitrary tensors of tags and values. But [`tf.compat.v1.summary.scalar`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/summary/scalar) requires a single scalar name and scalar value. In most cases, you can create [`tf.compat.v1.summary.scalar`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/summary/scalar) in a loop to get the same behavior

As before, TensorBoard groups charts by the top-level [`tf.name_scope`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/name_scope) which may
be inconvenient, for in the new summary ops, the summary will inherit that
[`tf.name_scope`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/name_scope) without user control. We plan to add more grouping mechanisms
to TensorBoard, so it will be possible to specify the TensorBoard group for
each summary via the summary API.

## Functions

[`audio_summary(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/audio_summary): Outputs a `Summary` protocol buffer with audio. (deprecated)

[`histogram_summary(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/histogram_summary): Outputs a `Summary` protocol buffer with a histogram. (deprecated)

[`image_summary(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/image_summary): Outputs a `Summary` protocol buffer with images. (deprecated)

[`merge_all_summaries(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/merge_all_summaries): Merges all summaries collected in the default graph. (deprecated)

[`merge_summary(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/merge_summary): Merges summaries. (deprecated)

[`scalar_summary(...)`](https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/contrib/deprecated/scalar_summary): Outputs a `Summary` protocol buffer with scalar values. (deprecated)