Sergio Guadarrama
Sergio Guadarrama
Thanks for your interested, we have debated a lot whether to allow expressions that need to be evaluated as gin arguments, but we have always came to the conclusion that...
Take a look at https://github.com/google/dopamine/blob/master/dopamine/agents/dqn/configs/dqn.gin for a real example.
I suppose if the `Class` has a property declaration for it I think it should work ``` @gin.register Class MyClass: def __init__(value): self._my_member_var = value @gin.register @property def my_member_var(self): return...
TensorFlow backend. Traceback (most recent call last): File "Adonia5.py", line 297, in gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=MEMORY_FRACTION) AttributeError: module 'tensorflow' has no attribute 'GPUOptions' C:\CarlaProject\PythonAPI\examples>python Adonia5.py Using TensorFlow backend. WARNING:tensorflow:From C:\Users\Hp\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow_core\python\compat\v2_compat.py:88: disable_resource_variables...
It seems an error in your code.
Yeah currently gin files doesn't support model aliases, since the import is just used to make sure the configurables are defined. But Gin allows to use shorter names if there...
You can just `import gin` and then create your own `external_configurables.py` for TF
Not sure if all the symbols would be defined in TF 1.8+
This a bit hard, since it will violate the Python behavior. But maybe Dan would have some insights in a few weeks.
Thanks @jackd if you want to make a PR we can review it and test it.