def test_explain(self, mock_post_request_func, mock_get_metadata,
mock_get_modality_map):
mock_post_request_func.return_value = {
'explanations': [{
'attributions_by_label': [{
'attributions': {
'data': [0.01, 0.02]
},
'baseline_score': 0.0001,
'example_score': 0.80,
'label_index': 3,
'output_name': 'probability'
}]
}]
}
m = ai_platform_model.AIPlatformModel('fake_end_point')
instances = [{'input': [0.05]}]
explanations = m.explain(instances)
self.assertTrue(mock_post_request_func.called)
tensor_dict = explanations[0].as_tensors()
self.assertTrue(
np.array_equal(tensor_dict['data'], np.asarray([0.01, 0.02])))
def test_predict(self):
self._mock_post_request_func.return_value = {'predictions': [0.5]}
m = ai_platform_model.AIPlatformModel('fake_end_point')
instances = [{'images_str': {'b64': u'fake_b64_str'}}]
predictions = m.predict(instances)
self.assertEqual(predictions['predictions'][0], 0.5)
def test_explain_attribution_key_error(self):
self._mock_post_request_func.return_value = {
'explanations': [{
'unknown_attribution_key': {}
}]
}
m = ai_platform_model.AIPlatformModel('fake_end_point')
with self.assertRaisesRegex(KeyError,
('Attribution keys are not present')):
m.explain([{'input': [0.05]}])
def test_explain_with_error(self, mock_post_request_func,
mock_get_metadata, mock_get_modality_map):
mock_post_request_func.return_value = {'error': 'This is an error.'}
m = ai_platform_model.AIPlatformModel('fake_end_point')
instances = [{'input': [0.05]}]
with self.assertRaisesRegex(
ValueError, ('Explanation call failed. .*\n'
'Original error message: "This is an error."')):
m.explain(instances)
def test_explain(self, mock_response):
self._mock_post_request_func.return_value = mock_response
m = ai_platform_model.AIPlatformModel(
'fake_end_point',
modality_to_inputs_map={constants.ALL_MODALITY: ['data']})
instances = [{'input': [0.05]}]
explanations = m.explain(instances)
self.assertTrue(self._mock_post_request_func.called)
tensor_dict = explanations[0].as_tensors()
self.assertTrue(
np.array_equal(tensor_dict['data'], np.asarray([0.01, 0.02])))
