Bug: Tests can fail in rare cases with random data

[cleong110]

Related to #146 possibly. Once in a while, tests like this one will fail, because the data used is randomly generated.

pose/src/python/tests/pose_test.py

Line 447 in e07ca68

def test_pose_tf_posebody_normalize_eager_mode_preserves_shape(self):

Possible solutions:

1. If we fix Minor bug: pose.normalize can divide by zero, filling data with nan values #146 it can fix this
2. Adjust the test data generation to ensure 0a and 0b are farther apart.

Example:

=================================== FAILURES ===================================
_ TestPoseTensorflowPoseBody.test_pose_tf_posebody_normalize_distribution_eager_mode_correct_result _

self = <tests.pose_test.TestPoseTensorflowPoseBody testMethod=test_pose_tf_posebody_normalize_distribution_eager_mode_correct_result>

    def test_pose_tf_posebody_normalize_distribution_eager_mode_correct_result(self):
        """
        Tests if the normalization of distribution for Pose object with TensorFlow PoseBody returns expected result.
        """
        pose = _get_random_pose_object_with_tf_posebody(num_keypoints=5, frames_min=2)
    
        tensor_as_numpy = pose.body.data.tensor.numpy()
        mask_as_numpy = pose.body.data.mask.numpy()
        masked_numpy_array = ma.array(tensor_as_numpy, mask=np.logical_not(mask_as_numpy))
    
        axis = (0, 1)
    
        numpy_mean = ma.mean(masked_numpy_array, axis=axis)
        numpy_std = ma.std(masked_numpy_array, axis=axis)
        expected_tensor = (masked_numpy_array - numpy_mean) / numpy_std
        expected_tensor = expected_tensor.filled(0)
    
        pose.normalize_distribution(axis=axis)
    
        actual_tensor_nan_fixed = pose.body.data.fix_nan()
        actual_tensor_zero_filled = actual_tensor_nan_fixed.zero_filled()
        actual_tensor_as_numpy = actual_tensor_zero_filled.numpy()
    
>       self.assertTrue(np.allclose(actual_tensor_as_numpy, expected_tensor),
                        "Normalize distribution did not return the expected result.")
E       AssertionError: False is not true : Normalize distribution did not return the expected result.

tests/pose_test.py:486: AssertionError
=============================== warnings summary ===============================
tests/hand_normalization_test.py::Test3DNormalization::test_hand_normalization
  /home/runner/work/pose/pose/src/python/pose_format/utils/normalization_3d.py:98: RuntimeWarning: invalid value encountered in divide
    normal /= np.linalg.norm(normal, axis=-1, keepdims=True)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=========================== short test summary info ============================
FAILED tests/pose_test.py::TestPoseTensorflowPoseBody::test_pose_tf_posebody_normalize_distribution_eager_mode_correct_result - AssertionError: False is not true : Normalize distribution did not return the expected result.
=================== 1 failed, 29 passed, 1 warning in 52.60s ===================