Fix numpy conversion for tf MirroredVariable.

[qlzh727]

This is a corner case that np.array requires the return type of array to be an array, and scalar will fail the type check.

This should address #18625

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (2ad8e07) 78.57% compared to head (2beed31) 57.73%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##           master   #18675       +/-   ##
===========================================
- Coverage   78.57%   57.73%   -20.85%     
===========================================
  Files         335      335               
  Lines       32979    32979               
  Branches     6455     6455               
===========================================
- Hits        25913    19039     -6874     
- Misses       5510    12518     +7008     
+ Partials     1556     1422      -134     

Flag	Coverage Δ
keras	57.73% <0.00%> (-20.75%)	⬇️
keras-jax	?
keras-numpy	57.73% <0.00%> (ø)
keras-tensorflow	?
keras-torch	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
keras/backend/common/variables.py	97.45% <0.00%> (-2.55%)	⬇️

... and 179 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fchollet]

Thanks for the PR! The test seems to be failing:

  File "/home/runner/work/keras/keras/integration_tests/tf_distribute_training_test.py", line 77, in <module>
    test_model_fit()
  File "/home/runner/work/keras/keras/integration_tests/tf_distribute_training_test.py", line 54, in test_model_fit
    history = model.fit(
  File "/home/runner/work/keras/keras/keras/utils/traceback_utils.py", line 123, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/home/runner/work/keras/keras/keras/backend/common/variables.py", line 186, in __array__
    return np.asarray(self.value, dtype)
ValueError: object __array__ method not producing an array

[qlzh727]

Thanks for the PR! The test seems to be failing:

  File "/home/runner/work/keras/keras/integration_tests/tf_distribute_training_test.py", line 77, in <module>
    test_model_fit()
  File "/home/runner/work/keras/keras/integration_tests/tf_distribute_training_test.py", line 54, in test_model_fit
    history = model.fit(
  File "/home/runner/work/keras/keras/keras/utils/traceback_utils.py", line 123, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/home/runner/work/keras/keras/keras/backend/common/variables.py", line 186, in __array__
    return np.asarray(self.value, dtype)
ValueError: object __array__ method not producing an array

ok. Now it should be fixed.

[fchollet]

LGTM, thank you