Fix the serialization bug of rectified adam. #1375
Conversation
|
You are owner of some files modified in this pull request. |
gabrieldemarmiesse
added
the
backport r0.8
| if isinstance(total_steps, float): | ||
| warnings.warn( | ||
| "The parameter `total_steps` passed to the __init__ of RectifiedAdam " | ||
| "is a float. This behavior is deprecated and in Addons 0.11, this " |
There was a problem hiding this comment.
Any reason to select 0.11? Also, how do we keep track of this when we get to version 0.11?
There was a problem hiding this comment.
We need to give a date in the warning, otherwise users won't care. They'll just say: "I'll leave tomorrow's probems to tomorrow's me". :P
No idea how to track it though. Milestones in github?
There was a problem hiding this comment.
Milestones require a due date, not version.
Why not provide a deprecation date instead of a version? But this directly means we'll need to create a new release on that due date.
There was a problem hiding this comment.
When stopping support for a branch, it's indeed common to specify a date: ex the python org will stop support for the branch 2.7 of python january 1st 2020.
When deprecating a feature and removing it later, it's more common to use a specific version:
/opt/conda/lib/python3.7/site-packages/tensorflow/python/training/tracking/data_structures.py:718: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated since Python 3.3,and in 3.9 it will stop working.
I believe it makes more sense to use a specific version. We can make a milestone. We're not forced to add a date. For example:
https://github.com/pytest-dev/pytest/milestones
There was a problem hiding this comment.
Sounds great! I'll accept this PR. We should create a milestone for this as well.
There was a problem hiding this comment.
Sure, I can do that.
There was a problem hiding this comment.
Haha, accidentally created a duplicate issue (#1381, closed now).
* Fix the serialization bug of rectified adam. * Better error message. * Update tensorflow_addons/optimizers/rectified_adam.py
Fix #1373