Fixed bug related to KerasSymbol.tensor #93
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
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Outdated suggestions cannot be applied.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.
This prevents the unfortunate scenario in which layer.set_weights() is called before training on some or all layers in a model, and then the model is trained and saved.
When the weights are set, since weight.tensor already exists from the random initialization, it is getting reassigned to a new symbol, whereas weight._bind_values[weight.name] remains pointing to the same symbol and only its values are being replaced by the new data. When model._sync_weights() is eventually called, the model._args[weight.name] and weight._bind_values[weight.name] contains the updated (trained) weights, but weight.tensor contains the old initialization values. And it is the weight.tensor which gets evaluated when calling model.save_weights() or layer.get_weights(). Therefore, incorrect weights are getting saved to the Keras model files.
Potentially, this is also the bug causing this issue.
The easy fix here is to keep the symbol which weight.tensor is pointing to and only replace it's values with new data, same as is done with weight._bind_values[weight.name].