|
|
@@ -0,0 +1,109 @@
|
|
|
+# -*- coding: utf-8 -*-
|
|
|
+"""verify 端点测试:L2→cosine 换算 + 阈值判定。"""
|
|
|
+import pytest
|
|
|
+from httpx import AsyncClient, ASGITransport
|
|
|
+from app.main import app
|
|
|
+
|
|
|
+
|
|
|
+class FakeDoc:
|
|
|
+ def __init__(self, content, metadata):
|
|
|
+ self.page_content = content
|
|
|
+ self.metadata = metadata
|
|
|
+
|
|
|
+
|
|
|
+class FakeCollection:
|
|
|
+ def __init__(self, emb_by_id):
|
|
|
+ self._emb = emb_by_id
|
|
|
+
|
|
|
+ def get(self, ids=None, include=None):
|
|
|
+ return {"ids": ids, "embeddings": [self._emb[i] for i in ids]}
|
|
|
+
|
|
|
+
|
|
|
+class FakeStore:
|
|
|
+ def __init__(self, emb_by_id, candidates):
|
|
|
+ self._collection = FakeCollection(emb_by_id)
|
|
|
+ self._candidates = candidates
|
|
|
+
|
|
|
+ def similarity_search_with_score(self, query, k=None, filter=None):
|
|
|
+ return self._candidates[:k]
|
|
|
+
|
|
|
+
|
|
|
+class FakeEmbeddings:
|
|
|
+ def __init__(self, q):
|
|
|
+ self.q = q
|
|
|
+
|
|
|
+ def embed_query(self, text):
|
|
|
+ return self.q
|
|
|
+
|
|
|
+
|
|
|
+def _patch_kb(monkeypatch, query_vec, emb_by_id, candidates):
|
|
|
+ import app.api.knowledge_base as kb
|
|
|
+ fake_store = FakeStore(emb_by_id, candidates)
|
|
|
+ monkeypatch.setattr(kb, "Chroma", lambda *a, **kw: fake_store)
|
|
|
+ monkeypatch.setattr(kb, "get_embeddings", lambda: FakeEmbeddings(query_vec))
|
|
|
+ return fake_store
|
|
|
+
|
|
|
+
|
|
|
+@pytest.mark.asyncio
|
|
|
+async def test_verify_converts_l2_to_cosine(monkeypatch):
|
|
|
+ """关键:L2 距离≠相似度。候选向量 [0.5,0.5,0,0] 与 query [1,0,0,0]
|
|
|
+ 真实 cosine=0.7071;若误用 L2(1-d^2/2) 会得 0.75。断言用真实 cosine。"""
|
|
|
+ q = [1.0, 0.0, 0.0, 0.0]
|
|
|
+ # 非归一化候选:真实 cos = 0.5 / sqrt(0.5) = 0.70710678
|
|
|
+ emb = {"cand1": [0.5, 0.5, 0.0, 0.0]}
|
|
|
+ cands = [(FakeDoc("AKK菌相关内容...", {"id": "cand1", "title": "AKK菌——下一代有益菌", "source": "microbiome"}), 0.7071)]
|
|
|
+ _patch_kb(monkeypatch, q, emb, cands)
|
|
|
+
|
|
|
+ transport = ASGITransport(app=app)
|
|
|
+ async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
|
+ resp = await client.post("/api/v1/knowledge-base/verify", json={
|
|
|
+ "sentences": [{"id": "s1", "text": "双歧杆菌有助于维持肠道菌群平衡"}],
|
|
|
+ "source": "microbiome", "k": 3, "threshold": 0.7,
|
|
|
+ })
|
|
|
+ assert resp.status_code == 200
|
|
|
+ data = resp.json()
|
|
|
+ assert data["code"] == 200
|
|
|
+ r = data["data"]["results"][0]
|
|
|
+ assert r["id"] == "s1"
|
|
|
+ got = r.get("max_similarity")
|
|
|
+ assert got == pytest.approx(0.70710678, abs=1e-3), (
|
|
|
+ "应为真实 cosine 约 0.7071,若返回约 0.75 说明误用 L2。got=" + str(got)
|
|
|
+ )
|
|
|
+ assert r["passed"] is True
|
|
|
+ assert r["matched_title"] == "AKK菌——下一代有益菌"
|
|
|
+
|
|
|
+
|
|
|
+@pytest.mark.asyncio
|
|
|
+async def test_verify_passed_when_cosine_ge_threshold(monkeypatch):
|
|
|
+ q = [1.0, 0.0, 0.0, 0.0]
|
|
|
+ emb = {"cand1": [0.9, 0.43589, 0.0, 0.0]} # norm≈1,cos=0.9 >=0.7
|
|
|
+ cands = [(FakeDoc("高相似内容", {"id": "cand1", "title": "T", "source": "microbiome"}), 0.3)]
|
|
|
+ _patch_kb(monkeypatch, q, emb, cands)
|
|
|
+
|
|
|
+ transport = ASGITransport(app=app)
|
|
|
+ async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
|
+ resp = await client.post("/api/v1/knowledge-base/verify", json={
|
|
|
+ "sentences": [{"id": "s1", "text": "双歧杆菌"}], "source": "microbiome",
|
|
|
+ })
|
|
|
+ data = resp.json()
|
|
|
+ r = data["data"]["results"][0]
|
|
|
+ assert r["passed"] is True
|
|
|
+ assert data["data"]["all_passed"] is True
|
|
|
+
|
|
|
+
|
|
|
+@pytest.mark.asyncio
|
|
|
+async def test_verify_failed_below_threshold(monkeypatch):
|
|
|
+ q = [1.0, 0.0, 0.0, 0.0]
|
|
|
+ emb = {"cand1": [0.3, 0.9, 0.0, 0.0]} # cos 约 0.316 < 0.7
|
|
|
+ cands = [(FakeDoc("低相似内容", {"id": "cand1", "title": "T", "source": "microbiome"}), 0.9)]
|
|
|
+ _patch_kb(monkeypatch, q, emb, cands)
|
|
|
+
|
|
|
+ transport = ASGITransport(app=app)
|
|
|
+ async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
|
+ resp = await client.post("/api/v1/knowledge-base/verify", json={
|
|
|
+ "sentences": [{"id": "s1", "text": "随便一句话"}], "source": "microbiome",
|
|
|
+ })
|
|
|
+ data = resp.json()
|
|
|
+ r = data["data"]["results"][0]
|
|
|
+ assert r["passed"] is False
|
|
|
+ assert data["data"]["all_passed"] is False
|