Set span property for ExampleAnomalies artifact in `DistributionVal…

…idator`. PiperOrigin-RevId: 628166134
tensorflow · Apr 30, 2024 · 0737d3c · 0737d3c
1 parent 81164a1
commit 0737d3c
Show file tree

Hide file tree

Showing 17 changed files with 148 additions and 1,983 deletions.
diff --git a/tfx/components/distribution_validator/executor.py b/tfx/components/distribution_validator/executor.py
@@ -366,6 +366,7 @@ def Do(
     anomalies_artifact.split_names = artifact_utils.encode_split_names(
         ['%s_%s' % (test, baseline) for test, baseline in split_pairs]
     )
+    anomalies_artifact.span = test_statistics.span
 
     validation_metrics_artifact = None
     if standard_component_specs.VALIDATION_METRICS_KEY in output_dict:

diff --git a/tfx/components/distribution_validator/executor_test.py b/tfx/components/distribution_validator/executor_test.py
@@ -220,11 +220,11 @@ def testSplitPairs(self, split_pairs, expected_split_pair_names,
                   component_generated_alert_list=[
                       component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                           alert_name=(
-                              '[train_eval][span 0] Feature-level anomalies '
+                              '[train_eval][span 2] Feature-level anomalies '
                               'present'
                           ),
                           alert_body=(
-                              '[train_eval][span 0] Feature(s) company, '
+                              '[train_eval][span 2] Feature(s) company, '
                               'dropoff_census_tract contain(s) anomalies. See '
                               'Anomalies artifact for more details.'
                           ),
@@ -260,11 +260,11 @@ def testSplitPairs(self, split_pairs, expected_split_pair_names,
                   component_generated_alert_list=[
                       component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                           alert_name=(
-                              '[train_eval][span 0] High num examples in '
+                              '[train_eval][span 2] High num examples in '
                               'current dataset versus the previous span.'
                           ),
                           alert_body=(
-                              '[train_eval][span 0] The ratio of num examples '
+                              '[train_eval][span 2] The ratio of num examples '
                               'in the current dataset versus the previous span '
                               'is 2.02094 (up to six significant digits), '
                               'which is above the threshold 1.'
@@ -372,11 +372,11 @@ def testSplitPairs(self, split_pairs, expected_split_pair_names,
                   component_generated_alert_list=[
                       component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                           alert_name=(
-                              '[train_eval][span 0] Feature-level anomalies '
+                              '[train_eval][span 2] Feature-level anomalies '
                               'present'
                           ),
                           alert_body=(
-                              '[train_eval][span 0] Feature(s) company '
+                              '[train_eval][span 2] Feature(s) company '
                               'contain(s) anomalies. See Anomalies artifact '
                               'for more details.'
                           ),
@@ -401,6 +401,7 @@ def testAnomaliesGenerated(
     stats_artifact.uri = os.path.join(source_data_dir, 'statistics_gen')
     stats_artifact.split_names = artifact_utils.encode_split_names(
         ['train', 'eval'])
+    stats_artifact.span = 2
 
     output_data_dir = os.path.join(
         os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', self.get_temp_dir()),
@@ -557,6 +558,7 @@ def testStructData(self):
     stats_artifact.split_names = artifact_utils.encode_split_names(
         ['train', 'eval']
     )
+    stats_artifact.span = 3
 
     struct_stats_train = text_format.Parse(
         """
@@ -684,9 +686,9 @@ def testStructData(self):
         component_generated_alert_list=[
             component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                 alert_name=(
-                    '[train_eval][span 0] Feature-level anomalies present'),
+                    '[train_eval][span 3] Feature-level anomalies present'),
                 alert_body=(
-                    '[train_eval][span 0] Feature(s) '
+                    '[train_eval][span 3] Feature(s) '
                     'parent_feature.value_feature contain(s) anomalies. See '
                     'Anomalies artifact for more details.'),
             )
@@ -1015,6 +1017,7 @@ def testEmptyData(self, stats_train, stats_eval, expected_anomalies):
     stats_artifact.uri = os.path.join(source_data_dir, 'statistics_gen')
     stats_artifact.split_names = artifact_utils.encode_split_names(
         ['train', 'eval'])
+    stats_artifact.span = 4
 
     validation_config = text_format.Parse(
         """
@@ -1100,10 +1103,10 @@ def testEmptyData(self, stats_train, stats_eval, expected_anomalies):
         component_generated_alert_list=[
             component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                 alert_name=(
-                    '[train_eval][span 0] Feature-level anomalies present'
+                    '[train_eval][span 4] Feature-level anomalies present'
                 ),
                 alert_body=(
-                    '[train_eval][span 0] Feature(s) first_feature contain(s) '
+                    '[train_eval][span 4] Feature(s) first_feature contain(s) '
                     'anomalies. See Anomalies artifact for more details.'
                 ),
             ),
@@ -1127,6 +1130,7 @@ def testAddOutput(self):
     stats_artifact.split_names = artifact_utils.encode_split_names(
         ['train', 'eval']
     )
+    stats_artifact.span = 5
 
     validation_config = text_format.Parse(
         """
@@ -1193,10 +1197,10 @@ def testAddOutput(self):
         component_generated_alert_list=[
             component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                 alert_name=(
-                    '[train_eval][span 0] Feature-level anomalies present'
+                    '[train_eval][span 5] Feature-level anomalies present'
                 ),
                 alert_body=(
-                    '[train_eval][span 0] Feature(s) '
+                    '[train_eval][span 5] Feature(s) '
                     'parent_feature.value_feature contain(s) anomalies. See '
                     'Anomalies artifact for more details.'
                 ),

diff --git a/tfx/components/example_validator/executor.py b/tfx/components/example_validator/executor.py
@@ -151,7 +151,6 @@ def Do(self, input_dict: Dict[str, List[types.Artifact]],
         output_dict[standard_component_specs.ANOMALIES_KEY])
     anomalies_artifact.split_names = artifact_utils.encode_split_names(
         split_names)
-    anomalies_artifact.span = stats_artifact.span
 
     schema = io_utils.SchemaReader().read(
         io_utils.get_only_uri_in_dir(

diff --git a/tfx/components/example_validator/executor_test.py b/tfx/components/example_validator/executor_test.py
@@ -160,15 +160,15 @@ def test_create_anomalies_alerts(self):
                       alert_name='Feature-level anomalies present',
                       alert_body=(
                           'Feature(s) company contain(s) anomalies for split '
-                          'train, span 11. See Anomalies artifact for more '
+                          'train, span 0. See Anomalies artifact for more '
                           'details.'
                       ),
                   ),
                   component_generated_alert_pb2.ComponentGeneratedAlertInfo(
                       alert_name='Feature-level anomalies present',
                       alert_body=(
                           'Feature(s) company contain(s) anomalies for split '
-                          'eval, span 11. See Anomalies artifact for more '
+                          'eval, span 0. See Anomalies artifact for more '
                           'details.'
                       ),
                   ),
@@ -190,7 +190,6 @@ def testDo(
     eval_stats_artifact.uri = os.path.join(source_data_dir, 'statistics_gen')
     eval_stats_artifact.split_names = artifact_utils.encode_split_names(
         ['train', 'eval', 'test'])
-    eval_stats_artifact.span = 11
 
     schema_artifact = standard_artifacts.Schema()
     schema_artifact.uri = os.path.join(source_data_dir, 'schema_gen')
@@ -232,7 +231,6 @@ def testDo(
     self.assertEqual(
         artifact_utils.encode_split_names(['train', 'eval']),
         validation_output.split_names)
-    self.assertEqual(eval_stats_artifact.span, validation_output.span)
 
     # Check example_validator outputs.
     train_anomalies_path = os.path.join(validation_output.uri, 'Split-train',