Cifar10 Example produces attribute error in TF compile call

[xchoo]

To replicate this error:

• Run the Cifar10 example

The error:

• The error is visible in the online docs, cell 8:

  Could not compute ANN values on this machine: in user code:
  
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:1224 test_function  *
      return step_function(self, iterator)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:1215 step_function  **
      outputs = model.distribute_strategy.run(run_step, args=(data,))
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:1211 run
      return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:2585 call_for_each_replica
      return self._call_for_each_replica(fn, args, kwargs)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:2945 _call_for_each_replica
      return fn(*args, **kwargs)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:1208 run_step  **
      outputs = model.test_step(data)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:1177 test_step
      y, y_pred, sample_weight, regularization_losses=self.losses)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/compile_utils.py:187 __call__
      self.build(y_pred)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/compile_utils.py:140 build
      self._losses = nest.map_structure(self._get_loss_object, self._losses)
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/util/nest.py:635 map_structure
      structure[0], [func(*x) for x in entries],
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/util/nest.py:635 <listcomp>
      structure[0], [func(*x) for x in entries],
  /home/travis/virtualenv/python3.6.7/lib/python3.6/site-packages/tensorflow/python/keras/engine/compile_utils.py:265 _get_loss_object
      loss_name = loss.__name__
  
  AttributeError: 'functools.partial' object has no attribute '__name__'

This issue was introduced in TF 2.2.0, and has been replicated in Python 3.6+. While it doesn't cause an issue for the docs build (because it uses pre-trained weights), this error will prevent users from training the model themselves.

This issue can be resolved by returning the exact function instance (instead of using partial), or by using a lambda function (but one of the code style checkers / formatters will complain).

[tbekolay]

Fixed in #315.