aufräumen des vector stores

math-tutor/backend/test/vector_store_pipeline_unit_test.py

+from __future__ import annotations
+
+import unittest
+
+from app.deterministic_services.vector_store import (
+    Retrieved,
+    Source,
+    SourceID,
+    build_default_pipeline_config,
+    expand_neighbor_children,
+    merge_retrieval_groups,
+    merge_sources,
+    select_dominant_scope,
+)
+
+
+def _mk_retrieved(
+    uid: str,
+    score: float,
+    chapter_index: int,
+    section_index: int,
+    subsection_index: int | None,
+    child_index: int = 1,
+) -> Retrieved:
+    return Retrieved(
+        uid=uid,
+        doc_type="child",
+        score=score,
+        metadata={
+            "chapter_index": chapter_index,
+            "section_index": section_index,
+            "subsection_index": subsection_index,
+            "child_index": child_index,
+            "chapter_title": "C",
+            "section_title": "S",
+            "subsection_title": "SS",
+            "title": uid,
+            "source_type": "child",
+            "path": "",
+            "doc_type": "child",
+            "uid": uid,
+        },
+        markdown=f"md-{uid}",
+    )
+
+
+class VectorStorePipelineUnitTest(unittest.TestCase):
+    def test_merge_sources_prefers_task_childs_on_duplicate(self) -> None:
+        child_direct = Source(
+            source_id=SourceID(
+                chapter_title="Kapitel",
+                section_title="Section",
+                subsection_title="Subsection",
+                title="Child A",
+                doc_type="child",
+            ),
+            retrieved_as="children_direct",
+            source_type="child",
+            score=0.7,
+            markdown="same-md",
+        )
+        task_child = Source(
+            source_id=SourceID(
+                chapter_title="Kapitel",
+                section_title="Section",
+                subsection_title="Subsection",
+                title="Child A",
+                doc_type="child",
+            ),
+            retrieved_as="task_childs",
+            source_type="child",
+            score=1.0,
+            markdown="same-md",
+        )
+
+        merged = merge_sources([child_direct], [task_child])
+
+        self.assertEqual(len(merged), 1)
+        self.assertEqual(merged[0].retrieved_as, "task_childs")
+
+    def test_build_default_pipeline_config(self) -> None:
+        cfg = build_default_pipeline_config(k=4, expand_links=True, neighbor_expand=2)
+        self.assertEqual(cfg.vector_k, 20)
+        self.assertEqual(cfg.scope_fill_k, 5)
+        self.assertTrue(cfg.enable_global_search)
+        self.assertTrue(cfg.enable_dominant_scope)
+        self.assertTrue(cfg.enable_scoped_child_search)
+        self.assertTrue(cfg.enable_context_docs)
+
+    def test_select_dominant_scope_prefers_count(self) -> None:
+        children = [
+            _mk_retrieved("a", 0.9, 1, 1, 1),
+            _mk_retrieved("b", 0.8, 1, 1, 1),
+            _mk_retrieved("c", 0.95, 1, 1, 2),
+        ]
+        scope = select_dominant_scope(children, level="subsection")
+        self.assertIsNotNone(scope)
+        assert scope is not None
+        self.assertEqual((scope.chapter_index, scope.section_index, scope.subsection_index), (1, 1, 1))
+
+    def test_select_dominant_scope_tiebreak_avg_score(self) -> None:
+        children = [
+            _mk_retrieved("a", 0.7, 1, 1, 1),
+            _mk_retrieved("b", 0.8, 1, 1, 2),
+        ]
+        scope = select_dominant_scope(children, level="subsection")
+        self.assertIsNotNone(scope)
+        assert scope is not None
+        self.assertEqual((scope.chapter_index, scope.section_index, scope.subsection_index), (1, 1, 2))
+
+    def test_select_dominant_scope_tiebreak_lexicographic(self) -> None:
+        children = [
+            _mk_retrieved("a", 0.8, 2, 1, 1),
+            _mk_retrieved("b", 0.8, 1, 2, 3),
+        ]
+        scope = select_dominant_scope(children, level="subsection")
+        self.assertIsNotNone(scope)
+        assert scope is not None
+        self.assertEqual((scope.chapter_index, scope.section_index, scope.subsection_index), (1, 2, 3))
+
+    def test_merge_retrieval_groups_dedup_max_score_and_trim_children(self) -> None:
+        groups = {
+            "children_direct": [
+                _mk_retrieved("u1", 0.5, 1, 1, 1),
+                _mk_retrieved("u2", 0.7, 1, 1, 1),
+            ],
+            "children_expanded": [
+                _mk_retrieved("u1", 0.9, 1, 1, 1),
+                _mk_retrieved("u3", 0.6, 1, 1, 1),
+            ],
+            "chapters": [],
+            "subsections": [],
+            "sections": [],
+            "neighbors": [_mk_retrieved("u2", 0.1, 1, 1, 1)],
+        }
+
+        merged = merge_retrieval_groups(groups, k=2)
+        self.assertEqual(sorted(item.uid for item in merged["children_direct"]), ["u1", "u2"])
+        score_u1 = [item.score for item in merged["children_direct"] if item.uid == "u1"][0]
+        self.assertEqual(score_u1, 0.9)
+        self.assertEqual(merged["children_expanded"], [])
+        self.assertEqual(merged["neighbors"], [])
+
+    def test_expand_neighbor_children_returns_empty_for_zero_expand(self) -> None:
+        children = [_mk_retrieved("u1", 0.8, 1, 1, 1, child_index=3)]
+        result = expand_neighbor_children("postgresql://unused", children, neighbor_expand=0)
+        self.assertEqual(result, [])
+
+
+if __name__ == "__main__":
+    unittest.main()