Closed Martin-Molinero closed 3 weeks ago
replacing build_output_shape.as_list() to list(build_output_shape) where as_list is used works as a fix, no other issues with keras 3 were observed after the change
replacing build_output_shape.as_list() to list(build_output_shape) where as_list is used works as a fix, no other issues with keras 3 were observed after the change
The modification solved my problem.
another problem with Keras 3 is when WeightNormalisation is enabled (use_weight_norm=True) because tensorflow_addons are not compatible with Keras 3.0 edit: https://github.com/tensorflow/addons master branch has fixes for import, but still there are problems to build ResidualBlock edit2: for Keras 3 exists alternative: https://keras.io/api/layers/normalization_layers/unit_normalization/ it also computes L2 norm of batch but scales it to 1
somehow managed to make WeightNormalization layer to run with Keras 3 and TCN . here are my attempts after which I saw TCN running https://github.com/tensorflow/addons/issues/2869
I tried doing that but i keep getting NameError: name 'build_output_shape' is not defined
I tried doing that but i keep getting NameError: name 'build_output_shape' is not defined
it should be self.build_output_shape as the variable is member of the class
Should be fixed with https://github.com/philipperemy/keras-tcn/pull/262
Running the following fails with keras 3.0.5
Error