Add Parametric Soft Exponential Unit (PSEU) activation layer #459
Conversation
Making keras-contrib compatible with tf.keras (keras-team#387)
|
@RaphaelMeudec and @gabrieldemarmiesse This is the new PR for PSEU (a copy of the previous one). Here, I have merged master into my branch and made it compatible with tf.keras. Any other changes ? |
|
@RaphaelMeudec and @gabrieldemarmiesse The build is now passing and the layer is tf.keras compatible. The ISRLU layer is also ready (build passing and tf.keras compatible) |
| self.trainable = False | ||
|
|
||
| if is_tf_keras: | ||
| def alpha_initializer(self, input_shape, dtype='float32', partition_info=None): |
There was a problem hiding this comment.
partition_info is a useless arg
There was a problem hiding this comment.
No need to separate is_tf_keras case
There was a problem hiding this comment.
@RaphaelMeudec The initializer does not work in tf.keras without the partition_info argument.
There was a problem hiding this comment.
Maybe you can use **kwargs to have only one declaration.
There was a problem hiding this comment.
@gabrieldemarmiesse Added **kwargs
|
@fchollet It would be great if you or someone else from your team can review this PR. |
Please make sure you've read and understood our contributing guidelines;
https://github.com/keras-team/keras-contrib/blob/master/CONTRIBUTING.md
-->
- What I did
See above
- How I did it
- How you can verify it
I added unit tests
This pull request fixes #issue_number_here