导入TensorFlow Federated时遭遇TypeError错误求助

尝试导入TensorFlow Federated（tff）并运行以下示例代码：

import collections
import numpy as np
import tensorflow as tf
import tensorflow_federated as tff

np.random.seed(0)

tff.federated_computation(lambda: 'Hello, World!')()

运行后触发如下TypeError错误：

/usr/local/lib/python3.8/dist-packages/tensorflow_federated/python/learning/metrics/keras_utils.py in <module>
 38     metrics_constructor: Union[MetricConstructor, MetricsConstructor,
 39                                MetricConstructors]
 ---> 40 ) -> Tuple[Callable[[], StateVar], Callable[[StateVar, ...], StateVar],
 41            Callable[[StateVar], Any]]:
 42   """Turn a Keras metric construction method into a tuple of pure functions.

/usr/lib/python3.8/typing.py in __getitem__(self, params)
814                                 f" Got {args}")
815             params = (tuple(args), result)
--> 816         return self.__getitem_inner__(params)
817 
818     @_tp_cache

/usr/lib/python3.8/typing.py in inner(*args, **kwds)
259         except TypeError:
260             pass  # All real errors (not unhashable args) are raised below.
--> 261         return func(*args, **kwds)
262     return inner
263 
/usr/lib/python3.8/typing.py in __getitem_inner__(self, params)
837                 return self.copy_with((_TypingEllipsis, result))
838             msg = "Callable[[arg, ...], result]: each arg must be a type."
--> 839             args = tuple(_type_check(arg, msg) for arg in args)
840             params = args + (result,)
841             return self.copy_with(params)

/usr/lib/python3.8/typing.py in <genexpr>(.0)
837                 return self.copy_with((_TypingEllipsis, result))
838             msg = "Callable[[arg, ...], result]: each arg must be a type."
---> 839             args = tuple(_type_check(arg, msg) for arg in args)
840             params = args + (result,)
841             return self.copy_with(params)

/usr/lib/python3.8/typing.py in _type_check(arg, msg, is_argument)
147         return arg
148     if not callable(arg):
--> 149         raise TypeError(f"{msg} Got {arg!r:.100}.")
150     return arg
151 

TypeError: Callable[[arg, ...], result]: each arg must be a type. Got Ellipsis.

解决方案 #

该错误源于Python 3.8的typing.Callable语法不支持在参数列表中直接使用省略号（...）表示可变参数，而新版TensorFlow Federated（TFF）使用了该语法，导致类型检查失败。

可通过以下两种方式解决：

• 升级Python版本：将Python升级至3.9或更高版本，新版本typing模块兼容Callable[[StateVar, ...], StateVar]这类语法。
• 降级TFF版本：安装适配Python 3.8的旧版TFF，例如tensorflow-federated==0.20.0，该版本的类型注解符合Python 3.8的规则。

执行降级的命令示例：

pip install tensorflow-federated==0.20.0

同时需确保TensorFlow版本与TFF版本匹配，避免出现其他兼容性问题。

内容的提问来源于stack exchange，提问作者Sully Marshall