Closed janstrohbeck closed 5 years ago
@janstrohbeck, well, wow, I need to think about it :smiley:
You don't have to merge that, I just wanted to push it so others can find it as well. It's really annoying when models can't be loaded for evaluation of a trained model. But this PR is just a workaround, no real solution.
@janstrohbeck, this fix can't get away from my head. Let's stick with this metaprogramming trick, and create an issue for future refactoring.
Sorry, I still have the problem with 1.1.0.
Hi @Gunkkk, can you, please, provide an example of the code that does not work to reproduce the problem?
Tricks keras upon serialization into saving only the name of the metric to be stored, which allows loading the model sucessfully. See #13