Port ProximalAdagrad #1976
Port ProximalAdagrad #1976
Conversation
| super().__init__(name, **kwargs) | ||
| self._set_hyper("learning_rate", kwargs.get("lr", learning_rate)) | ||
| self.l1_regularization_strength = l1_regularization_strength | ||
| self.l2_regularization_strength = l2_regularization_strength |
There was a problem hiding this comment.
How about creating l1_regularxx and l2_regularxx as hyper-parameter?
There was a problem hiding this comment.
Will do. Just copy paste the name from compat.v1. Maybe l1 and l2 are more suitable?
https://www.tensorflow.org/api_docs/python/tf/keras/regularizers/L1L2
There was a problem hiding this comment.
I agree, l1 and l2 are more concise :-)
There was a problem hiding this comment.
Tzu-Wei, perhaps we could use self._set_hyper for l1 and l2, what do you think?
There was a problem hiding this comment.
of course. let me update it real quick :-)
There was a problem hiding this comment.
Seems that _set_hyper/_get_hyper supports more types of inputs.
|
Umm, sorry for the confusion, but |
* Port ProximalAdagrad optimizer * Update code owner * Address comments * Fix tests * run formatter * Reorder import * Use set_hyper/get_hyper * Change back to *_regularization_stength
* Port ProximalAdagrad optimizer * Update code owner * Address comments * Fix tests * run formatter * Reorder import * Use set_hyper/get_hyper * Change back to *_regularization_stength
Closes #591.