Closed theRealSuperMario closed 4 years ago
please provide a plan of action: There is not much edflow can do about the way Keras handles attributes, so imho the best we can do is mention it in the faq. Can you add an entry for it and then close this issue?
Adding it to FAQ is a good idea.
documented in #252
When adding the config to a model that inherits from
tf.keras.Model
, the config cannot be dumped. It looks like keras changes lists within the config to aListWrapper
object, which are not reduceably byyaml.dump
To Reproduce
run
edflow -b config.yaml -t -n foo -d
givesExpected behavior I expect the config to be dumped
Additional context Workaround is to simply not do
self.config = config
and save everything you need in a field in the model.