Support Keras APIs and fix save-restore issue in TF2 #119
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Lifann
changed the title
rhdong
reviewed
Jul 20, 2021
tensorflow_recommenders_addons/dynamic_embedding/python/ops/cuckoo_hashtable_ops.py
Show resolved
Hide resolved
rhdong
reviewed
Aug 5, 2021
| @@ -771,7 +816,8 @@ def safe_embedding_lookup_sparse( | |||
| entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean" the | |||
| default. | |||
| default_id: The id to use for an entry with no features. | |||
| name: A name for this operation (optional). | |||
| name: A name for this operation. Name is optional in graph mode and required | |||
There was a problem hiding this comment.
How to confirm the name is set in eager mode?
rhdong
reviewed
Aug 5, 2021
|
|
||
| # TODO(Lifann) Use more resonable name. | ||
| def _create_trainable(trainable_name): | ||
| return de.TrainableWrapper(params, |
There was a problem hiding this comment.
Plz check here , there is assume that 'TrainableWrapper' is always appear in the TW name, and the placer won't place TW onto PSs for multi-thread safe . device placement
There was a problem hiding this comment.
Accept. Use ops.colocate_with to force the TW to reside with ids.
rhdong
reviewed
Aug 5, 2021
| @@ -81,10 +82,11 @@ def __init__(self, params, ids, max_norm, *args, **kwargs): | |||
| self.prefetch_values_op = None | |||
| self.model_mode = kwargs.get("model_mode") | |||
| kwargs.pop("model_mode") | |||
| self._slots_track = [] | |||
There was a problem hiding this comment.
_slots_tracked or _tracked_slots maybe better.
rhdong
reviewed
Aug 5, 2021
| @@ -443,7 +464,8 @@ def embedding_lookup( | |||
| params: A dynamic_embedding.Variable instance. | |||
| ids: A tensor with any shape as same dtype of params.key_dtype. | |||
| partition_strategy: No used, for API compatiblity with `nn.emedding_lookup`. | |||
| name: A name for the operation (optional). | |||
| name: A name for the operation. Name is optional in graph mode and required | |||
| in eager mode. | |||
There was a problem hiding this comment.
I think a check is needed to make sure the name is set in eager mode.
There was a problem hiding this comment.
It's checked when creating TW. I moved it to the front of func entry now.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR makes Keras APIs become compatible to TFRA, and also fixes the save/restore issue in TensorFlow 2.x version.
Thanks to @ccsquare 's suggestions on save/restore.
Main Features
1. Support object-based programming
2. Add a demo on amazon digital video games in object-based APIs.
3. Provide fine-grained API to access non-explicitly defined
dynamic_embedding.Variableand deprecatedynamic_embedding.GraphKeys.4. Fix relative bugs: