Cyclical learning rates broken in TFA 0.8.2 #1158
Description
System information
- OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
Windows 10.0.16299 Build 16299 - TensorFlow version and how it was installed (source or binary):
2.1.0 installed via pip - TensorFlow-Addons version and how it was installed (source or binary):
0.8.2 installed via pip - Python version:
3.6.8 - Is GPU used? (yes/no):
no
Code to reproduce the issue
from tensorflow_addons.optimizers import cyclical_learning_rate
callback = cyclical_learning_rate.ExponentialCyclicalLearningRate(
initial_learning_rate=1e-8,
maximal_learning_rate=1e-2,
step_size=100,
)
Other info / logs
Stack trace:
Traceback (most recent call last):
File "tfa_error.py", line 6, in <module>
step_size=100,
File "C:\dev\anaconda3\envs\lamro\lib\site-packages\typeguard\__init__.py", line 808, in wrapper
memo = _CallMemo(python_func, _localns, args=args, kwargs=kwargs)
File "C:\dev\anaconda3\envs\lamro\lib\site-packages\typeguard\__init__.py", line 93, in __init__
self.arguments = self.signature.bind(*args, **kwargs).arguments
File "C:\dev\anaconda3\envs\lamro\lib\inspect.py", line 2997, in bind
return args[0]._bind(args[1:], kwargs)
File "C:\dev\anaconda3\envs\lamro\lib\inspect.py", line 2968, in _bind
format(arg=param_name)) from None
TypeError: missing a required argument: 'scale_fn'
The issue was caused by commit f42ca1dd7487c237b8adbc5014d5a54af5c7e4e2 which added type hints to the cyclic learning rates module. An additional scale_fn argument was copied over from the base CyclicalLearningRate class to its subclasses despite the subclasses not needing that argument.
