Dateien nach "tests" hochladen

tests/test_retriever_weight.py

@@ -0,0 +1,48 @@
+from app.core import retriever as r
+from app.models.dto import QueryRequest
+import app.core.qdrant as qdr
+import app.core.qdrant_points as qp
+import app.services.embeddings_client as ec
+
+
+class _DummyClient:
+    pass
+
+
+def _fake_get_client(cfg):
+    return _DummyClient()
+
+
+def _fake_embed_text(text: str):
+    # Dimension egal, da wir search_chunks_by_vector faken
+    return [0.0] * 384
+
+
+def _fake_search_chunks_by_vector_weights(client, prefix, vector, top=10, filters=None):
+    # Zwei Treffer mit unterschiedlichem retriever_weight
+    return [
+        ("chunk:low", 0.9, {"note_id": "note:low", "path": "low.md", "section_title": "S1", "retriever_weight": 0.2}),
+        ("chunk:high", 0.8, {"note_id": "note:high", "path": "high.md", "section_title": "S2", "retriever_weight": 1.5}),
+    ]
+
+
+def test_retriever_weight_influences_ranking(monkeypatch):
+    # Konfiguration: Qdrant & Embedding faken
+    monkeypatch.setattr(qdr, "get_client", _fake_get_client)
+    monkeypatch.setattr(ec, "embed_text", _fake_embed_text)
+    monkeypatch.setattr(qp, "search_chunks_by_vector", _fake_search_chunks_by_vector_weights)
+
+    req = QueryRequest(mode="semantic", query="karate", top_k=2)
+    resp = r.semantic_retrieve(req)
+
+    assert len(resp.results) == 2
+
+    first = resp.results[0]
+    second = resp.results[1]
+
+    # Semantic Scores: 0.9 (low), 0.8 (high)
+    # Total Scores: 0.9 * 0.2 = 0.18 vs. 0.8 * 1.5 = 1.2 -> high sollte oben liegen
+    assert first.note_id == "note:high"
+    assert second.note_id == "note:low"
+
+    assert first.total_score > second.total_score