Fix Recommendations AI catalog outputs and sklearn pins for 3.14

Author: aIbrahiim

sdks/python/apache_beam/examples/inference/sklearn_examples_requirements.txt

@@ -20,4 +20,5 @@
 # However, newer sklearn is needed for testing on newer Python version
 scikit-learn==1.0.2; python_version < '3.11'
 # bump sklearn version when new Python version is supported
-scikit-learn==1.7.1; python_version >= '3.11'
+scikit-learn==1.7.1; python_version >= '3.11' and python_version < '3.14'
+scikit-learn==1.7.2; python_version >= '3.14'

sdks/python/apache_beam/ml/gcp/recommendations_ai.py

@@ -79,15 +79,19 @@ def get_recommendation_user_event_client():
 
 
 class CreateCatalogItem(PTransform):
-  """Creates catalogitem information.
-    The ``PTransform`` returns a PCollectionTuple with a PCollections of
-    successfully and failed created CatalogItems.
+  """Creates catalog item records.
+
+    The ``PTransform`` returns a ``PCollectionTuple`` of successfully created
+    catalog items (``created_catalog_items``) and failures
+    (``failed_catalog_items``).
 
     Example usage::
 
-      pipeline | CreateCatalogItem(
-        project='example-gcp-project',
-        catalog_name='my-catalog')
+      result = (
+          pipeline
+          | CreateCatalogItem(
+              project='example-gcp-project', catalog_name='my-catalog'))
+      created = result.created_catalog_items
     """
   def __init__(
       self,
@@ -123,13 +127,15 @@ def expand(self, pcoll):
       raise ValueError(
           """GCP project name needs to be specified in "project" pipeline
             option""")
-    return pcoll | ParDo(
+    pardo = ParDo(
         _CreateCatalogItemFn(
             self.project,
             self.retry,
             self.timeout,
             self.metadata,
             self.catalog_name))
+    return pcoll | pardo.with_outputs(
+        FAILED_CATALOG_ITEMS, main='created_catalog_items')
 
 
 class _CreateCatalogItemFn(DoFn):

sdks/python/apache_beam/ml/gcp/recommendations_ai_test.py

@@ -65,10 +65,12 @@ def test_CreateCatalogItem(self):
                            return_value=self._mock_client):
       p = beam.Pipeline()
 
-      _ = (
+      create_outputs = (
           p | "Create data" >> beam.Create([self._catalog_item])
           | "Create CatalogItem" >>
           recommendations_ai.CreateCatalogItem(project="test"))
+      _ = create_outputs.created_catalog_items | beam.combiners.Count.Globally()
+      _ = create_outputs.failed_catalog_items | beam.combiners.Count.Globally()
 
       result = p.run()
       result.wait_until_finish()

sdks/python/apache_beam/ml/gcp/recommendations_ai_test_it.py

@@ -77,11 +77,14 @@ def test_create_catalog_item(self):
 
     with TestPipeline(is_integration_test=True) as p:
       RecommendationAIIT.test_ran = True
-      output = (
+      create_outputs = (
           p | 'Create data' >> beam.Create([CATALOG_ITEM])
           | 'Create CatalogItem' >>
-          recommendations_ai.CreateCatalogItem(project=GCP_TEST_PROJECT)
+          recommendations_ai.CreateCatalogItem(project=GCP_TEST_PROJECT))
+      output = (
+          create_outputs.created_catalog_items
           | beam.ParDo(extract_id) | beam.combiners.ToList())
+      _ = create_outputs.failed_catalog_items | beam.combiners.Count.Globally()
 
       assert_that(output, equal_to([[CATALOG_ITEM["id"]]]))