Docstring format conformity, can you use Google's format?

[SamuelMarks]

Happy to send the PR(s).

Although there isn't a precise grammar, there is:

• https://google.github.io/styleguide/pyguide.html#doc-function-args @ ddccc0f
• https://chromium.googlesource.com/chromiumos/docs/+/master/styleguide/python.md#describing-arguments-in-docstrings @ 9fc0fc0
• https://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html @ c0ae8e3

Taken from that last link:

        param2 (:obj:`str`, optional): The second parameter. Defaults to None.

Such a PR would involve these kind of patches:

diff --git a/keras/utils/audio_dataset_utils.py b/keras/utils/audio_dataset_utils.py
index a60784380..4c00de4e2 100644
--- a/keras/utils/audio_dataset_utils.py
+++ b/keras/utils/audio_dataset_utils.py
@@ -85,7 +85,7 @@ def audio_dataset_from_directory(
             If set to `False`, sorts the data in alphanumeric order.
             Defaults to `True`.
         seed: Optional random seed for shuffling and transformations.
-        validation_split: Optional float between 0 and 1, fraction of data to
+        validation_split (`float`, optional): Value between 0 and 1, fraction of data to
             reserve for validation.
         subset: Subset of the data to return. One of `"training"`,
             `"validation"` or `"both"`. Only used if `validation_split` is set.

Is this something you would be interested in switching to?

(this would make parsing your codebase unambiguous for both developer and code parsers)

Thanks for your consideration

[hertschuh]

@SamuelMarks

Question: the codebase does not have type annotations, how would you generate such PRs?
The main issue (which is why we haven't added type annotations) is that many parameters are complex unions.

[SamuelMarks]

@hertschuh I can automatically apply type annotations throughout the whole codebase using my open-source library.

If you prefer them to be in the docstrings, like we did pre Python 3.6, then my library also supports that syntax (and NumPy, Google, and ReST docstring formats).

Will need some manually massaging beforehand though, as there is textual ambiguity throughout the documentation.

Anyway I can very quickly fix the docstrings if you are open to the PRs. Want a PR for all of keras, per major module, or per file?

(can send them all before christmas 🎄 )