test_WP22_intelligence aktualisiert roundtrip
This commit is contained in:
parent
7f7d8c87db
commit
342d3e5103
|
|
@ -1,109 +1,227 @@
|
|||
"""
|
||||
FILE: tests/test_WP22_integration.py
|
||||
DESCRIPTION: Integrationstest für WP-22 (Graph Intelligence).
|
||||
Prüft: Registry, Lifecycle Scoring, Router-Logik und Regression.
|
||||
Mockt Datenbank und LLM, um Logikfehler isoliert zu finden.
|
||||
"""
|
||||
import unittest
|
||||
import os
|
||||
import shutil
|
||||
import json
|
||||
from unittest.mock import patch
|
||||
import yaml
|
||||
from unittest.mock import MagicMock, patch, AsyncMock
|
||||
from datetime import datetime
|
||||
|
||||
# --- FIX: Import der KORREKTEN Funktion und Klassen ---
|
||||
# --- Imports der App-Module ---
|
||||
# Wir gehen davon aus, dass wir im Root-Verzeichnis sind.
|
||||
from app.models.dto import ChatRequest, QueryRequest, QueryHit
|
||||
from app.services.edge_registry import EdgeRegistry
|
||||
from app.core.retriever import _compute_total_score, _get_status_multiplier
|
||||
from app.routers.chat import _classify_intent, get_decision_strategy, chat_endpoint
|
||||
|
||||
class TestWP22Intelligence(unittest.TestCase):
|
||||
# --- Test Suite ---
|
||||
class TestWP22Integration(unittest.TestCase):
|
||||
|
||||
def setUp(self):
|
||||
# 1. Test-Vault Struktur definieren
|
||||
self.test_vault_root = os.path.abspath("tests/temp_vault")
|
||||
self.user_manual_dir = os.path.join(self.test_vault_root, "01_User_Manual")
|
||||
self.log_dir = os.path.abspath("tests/logs")
|
||||
"""Bereitet eine isolierte Test-Umgebung vor."""
|
||||
self.test_dir = "tests/temp_integration"
|
||||
self.os_env_patch = patch.dict(os.environ, {
|
||||
"MINDNET_VAULT_ROOT": self.test_dir,
|
||||
"MINDNET_DECISION_CONFIG": os.path.join(self.test_dir, "config", "decision_engine.yaml"),
|
||||
"MINDNET_TYPES_FILE": os.path.join(self.test_dir, "config", "types.yaml")
|
||||
})
|
||||
self.os_env_patch.start()
|
||||
|
||||
# 2. Verzeichnisse erstellen
|
||||
os.makedirs(self.user_manual_dir, exist_ok=True)
|
||||
os.makedirs(self.log_dir, exist_ok=True)
|
||||
# Verzeichnisse erstellen
|
||||
os.makedirs(os.path.join(self.test_dir, "config"), exist_ok=True)
|
||||
os.makedirs(os.path.join(self.test_dir, "01_User_Manual"), exist_ok=True)
|
||||
os.makedirs(os.path.join(self.test_dir, "data", "logs"), exist_ok=True)
|
||||
|
||||
# 3. Dummy Vocabulary File am korrekten Ort erstellen
|
||||
# Der Ort muss exakt '01_User_Manual/01_edge_vocabulary.md' relativ zum vault_root sein
|
||||
self.vocab_file = os.path.join(self.user_manual_dir, "01_edge_vocabulary.md")
|
||||
with open(self.vocab_file, "w", encoding="utf-8") as f:
|
||||
f.write("""
|
||||
| **canonical** | Aliases |
|
||||
| :--- | :--- |
|
||||
| **caused_by** | ursache_ist, wegen |
|
||||
| **next** | danach, folgt |
|
||||
""")
|
||||
# 1. Config: decision_engine.yaml (mit Boosts)
|
||||
self.decision_config = {
|
||||
"strategies": {
|
||||
"FACT": {
|
||||
"trigger_keywords": ["was ist"],
|
||||
"edge_boosts": {"part_of": 2.0}
|
||||
},
|
||||
"CAUSAL": {
|
||||
"trigger_keywords": ["warum", "weshalb"],
|
||||
"edge_boosts": {"caused_by": 3.0, "related_to": 0.5}
|
||||
}
|
||||
}
|
||||
}
|
||||
with open(os.environ["MINDNET_DECISION_CONFIG"], "w") as f:
|
||||
yaml.dump(self.decision_config, f)
|
||||
|
||||
# 4. Registry Reset & Init mit explizitem Vault Root
|
||||
# Wir setzen das Singleton zurück, um sicherzustellen, dass es unseren Pfad nutzt
|
||||
# 2. Config: Edge Vocabulary (für Registry)
|
||||
with open(os.path.join(self.test_dir, "01_User_Manual", "01_edge_vocabulary.md"), "w") as f:
|
||||
f.write("| **caused_by** | ursache_ist, wegen |\n| **part_of** | teil_von |")
|
||||
|
||||
# 3. Registry Reset
|
||||
EdgeRegistry._instance = None
|
||||
# Hier übergeben wir den Test-Vault-Pfad! Das Skript ignoriert jetzt die .env für den Test.
|
||||
self.registry = EdgeRegistry(vault_root=self.test_vault_root)
|
||||
self.registry.unknown_log_path = os.path.join(self.log_dir, "unknown_edges.jsonl")
|
||||
self.registry = EdgeRegistry(vault_root=self.test_dir)
|
||||
|
||||
def tearDown(self):
|
||||
if os.path.exists(self.test_vault_root):
|
||||
shutil.rmtree(self.test_vault_root)
|
||||
if os.path.exists("tests/logs"):
|
||||
shutil.rmtree("tests/logs")
|
||||
self.os_env_patch.stop()
|
||||
if os.path.exists(self.test_dir):
|
||||
shutil.rmtree(self.test_dir)
|
||||
EdgeRegistry._instance = None
|
||||
|
||||
def test_registry_resolution(self):
|
||||
print("\n--- Test A: Registry & Alias Resolution ---")
|
||||
# ------------------------------------------------------------------------
|
||||
# TEST 1: Edge Registry & Validation (WP-22 B)
|
||||
# ------------------------------------------------------------------------
|
||||
def test_edge_registry_aliases(self):
|
||||
print("\n🔵 TEST 1: Edge Registry Resolution")
|
||||
# Test: Alias Auflösung
|
||||
resolved = self.registry.resolve("ursache_ist")
|
||||
self.assertEqual(resolved, "caused_by", "Alias 'ursache_ist' sollte zu 'caused_by' werden.")
|
||||
|
||||
# Prüfen ob Pfad korrekt übernommen wurde
|
||||
expected_path = os.path.join(self.test_vault_root, "01_User_Manual", "01_edge_vocabulary.md")
|
||||
# Da wir abspath nutzen, vergleichen wir normalized paths
|
||||
self.assertTrue(os.path.exists(expected_path), "Test fixture file was not created correctly")
|
||||
# Test: Unknown Logging
|
||||
unknown = self.registry.resolve("foobar_link")
|
||||
self.assertEqual(unknown, "foobar_link", "Unbekannte Kanten sollen durchgereicht werden.")
|
||||
|
||||
if not self.registry.valid_types:
|
||||
self.fail(f"Registry empty! Root used: {self.registry.vault_root}")
|
||||
|
||||
self.assertEqual(self.registry.resolve("caused_by"), "caused_by")
|
||||
self.assertEqual(self.registry.resolve("ursache_ist"), "caused_by")
|
||||
|
||||
unknown = self.registry.resolve("mystery_link")
|
||||
self.assertEqual(unknown, "mystery_link")
|
||||
|
||||
# Prüfen ob Logging funktioniert
|
||||
if os.path.exists(self.registry.unknown_log_path):
|
||||
with open(self.registry.unknown_log_path, "r") as f:
|
||||
self.assertIn("mystery_link", f.read())
|
||||
print("✅ Registry loaded from custom vault root & validated.")
|
||||
else:
|
||||
self.fail("Logfile was not created.")
|
||||
|
||||
def test_lifecycle_scoring(self):
|
||||
print("\n--- Test B: Lifecycle Scoring Math ---")
|
||||
base_sem = 0.9
|
||||
payload_draft = {"status": "draft"}
|
||||
payload_stable = {"status": "stable"}
|
||||
log_path = self.registry.unknown_log_path
|
||||
self.assertTrue(os.path.exists(log_path), "Logfile für unbekannte Kanten fehlt.")
|
||||
with open(log_path, "r") as f:
|
||||
self.assertIn("foobar_link", f.read())
|
||||
print("✅ Registry funktioniert (Alias + Logging).")
|
||||
|
||||
# ------------------------------------------------------------------------
|
||||
# TEST 2: Lifecycle Scoring (WP-22 A)
|
||||
# ------------------------------------------------------------------------
|
||||
def test_lifecycle_scoring_logic(self):
|
||||
print("\n🔵 TEST 2: Lifecycle Scoring (Draft vs. Stable)")
|
||||
# Mock Weights: Sem=1.0, Edge=0.5, Cent=0.0
|
||||
with patch("app.core.retriever._get_scoring_weights", return_value=(1.0, 0.5, 0.0)):
|
||||
base_sem = 0.9
|
||||
|
||||
# Case A: Draft (Malus)
|
||||
payload_draft = {"status": "draft", "retriever_weight": 1.0}
|
||||
mult_draft = _get_status_multiplier(payload_draft)
|
||||
self.assertEqual(mult_draft, 0.8, "Draft sollte 0.8 Multiplier haben.")
|
||||
|
||||
score_draft, _, _ = _compute_total_score(base_sem, payload_draft)
|
||||
|
||||
# Case B: Stable (Bonus)
|
||||
payload_stable = {"status": "stable", "retriever_weight": 1.0}
|
||||
mult_stable = _get_status_multiplier(payload_stable)
|
||||
self.assertEqual(mult_stable, 1.2, "Stable sollte 1.2 Multiplier haben.")
|
||||
|
||||
score_draft = base_sem * mult_draft
|
||||
score_stable = base_sem * mult_stable
|
||||
score_stable, _, _ = _compute_total_score(base_sem, payload_stable)
|
||||
|
||||
self.assertLess(score_draft, base_sem)
|
||||
self.assertGreater(score_stable, base_sem)
|
||||
print("✅ Lifecycle scoring math verified.")
|
||||
print(f" Draft Score: {score_draft:.2f} | Stable Score: {score_stable:.2f}")
|
||||
self.assertGreater(score_stable, score_draft)
|
||||
print("✅ Lifecycle Scoring korrekt implementiert.")
|
||||
|
||||
def test_dynamic_boosting(self):
|
||||
print("\n--- Test C: Dynamic Edge Boosting ---")
|
||||
semantic_score = 0.5
|
||||
raw_edge_bonus = 1.0
|
||||
payload = {"status": "active"}
|
||||
# ------------------------------------------------------------------------
|
||||
# TEST 3: Semantic Router & Boosting (WP-22 C)
|
||||
# ------------------------------------------------------------------------
|
||||
async def test_router_integration(self):
|
||||
print("\n🔵 TEST 3: Semantic Router Integration")
|
||||
|
||||
with patch("app.core.retriever._get_scoring_weights", return_value=(1.0, 1.0, 0.0)):
|
||||
score_normal, _, _ = _compute_total_score(
|
||||
semantic_score, payload, edge_bonus=raw_edge_bonus, dynamic_edge_boosts=None
|
||||
)
|
||||
boost_map = {"caused_by": 2.0}
|
||||
score_boosted, _, _ = _compute_total_score(
|
||||
semantic_score, payload, edge_bonus=raw_edge_bonus, dynamic_edge_boosts=boost_map
|
||||
# Mock LLM Service (für Fallback, wird hier aber durch Keywords umgangen)
|
||||
mock_llm = MagicMock()
|
||||
mock_llm.prompts = {}
|
||||
|
||||
# --- Szenario A: Kausal-Frage ("Warum...") ---
|
||||
query_causal = "Warum ist das Projekt gescheitert?"
|
||||
|
||||
# 1. Intent Detection prüfen
|
||||
intent, source = await _classify_intent(query_causal, mock_llm)
|
||||
self.assertEqual(intent, "CAUSAL", "Sollte 'CAUSAL' Intent erkennen via Keywords.")
|
||||
|
||||
# 2. Strategy Load prüfen
|
||||
strategy = get_decision_strategy(intent)
|
||||
boosts = strategy.get("edge_boosts", {})
|
||||
self.assertEqual(boosts.get("caused_by"), 3.0, "Sollte 'caused_by' Boost von 3.0 laden.")
|
||||
|
||||
print(f" Intent: {intent} -> Boosts: {boosts}")
|
||||
print("✅ Router lädt Config korrekt.")
|
||||
|
||||
# ------------------------------------------------------------------------
|
||||
# TEST 4: Full Pipeline (Chat -> Retriever)
|
||||
# ------------------------------------------------------------------------
|
||||
async def test_full_pipeline_flow(self):
|
||||
print("\n🔵 TEST 4: Full Chat Pipeline (Integration)")
|
||||
|
||||
# Mocks
|
||||
mock_llm = AsyncMock()
|
||||
mock_llm.prompts = {}
|
||||
mock_llm.generate_raw_response.return_value = "Das ist die Antwort."
|
||||
|
||||
mock_retriever = AsyncMock()
|
||||
# Mock Search Result
|
||||
mock_hit = QueryHit(
|
||||
node_id="123", semantic_score=0.9, edge_bonus=0.5, centrality_bonus=0.0, total_score=1.0,
|
||||
source={"text": "Inhalt"}, payload={"type": "concept"}
|
||||
)
|
||||
mock_retriever.search.return_value.results = [mock_hit]
|
||||
|
||||
self.assertGreater(score_boosted, score_normal)
|
||||
print("✅ Dynamic Boosting logic verified.")
|
||||
# Request: "Warum..."
|
||||
req = ChatRequest(message="Warum ist das passiert?", top_k=3)
|
||||
|
||||
# EXECUTE Endpoint
|
||||
# Wir müssen sicherstellen, dass _load_decision_config unsere Test-Config nutzt.
|
||||
# Da wir os.environ gepatcht haben, sollte das klappen.
|
||||
|
||||
# Wir müssen die Caches leeren, da Module-Level Variablen sonst alte Werte haben
|
||||
import app.routers.chat
|
||||
app.routers.chat._DECISION_CONFIG_CACHE = None
|
||||
|
||||
response = await chat_endpoint(req, llm=mock_llm, retriever=mock_retriever)
|
||||
|
||||
# ASSERTIONS
|
||||
|
||||
# 1. Wurde der Retriever mit den Boosts aufgerufen?
|
||||
# Wir inspecten das Argument 'boost_edges' im call_args des Retrievers
|
||||
called_query_req = mock_retriever.search.call_args[0][0]
|
||||
|
||||
self.assertIsNotNone(called_query_req.boost_edges, "Retriever sollte boost_edges erhalten.")
|
||||
self.assertEqual(called_query_req.boost_edges.get("caused_by"), 3.0, "Boost 'caused_by' sollte 3.0 sein.")
|
||||
|
||||
# 2. Wurde der Intent korrekt durchgereicht?
|
||||
self.assertEqual(response.intent, "CAUSAL")
|
||||
|
||||
print(f" Retriever called with: {called_query_req.boost_edges}")
|
||||
print("✅ Pipeline reicht Boosts erfolgreich weiter.")
|
||||
|
||||
# ------------------------------------------------------------------------
|
||||
# TEST 5: Regression Check (Fallback behavior)
|
||||
# ------------------------------------------------------------------------
|
||||
async def test_regression_standard_query(self):
|
||||
print("\n🔵 TEST 5: Regression (Standard Query)")
|
||||
|
||||
# Request ohne Keyword -> Sollte FACT (Default) sein
|
||||
# Oder LLM Fallback (hier gemockt)
|
||||
|
||||
mock_llm = AsyncMock()
|
||||
mock_llm.prompts = {}
|
||||
# Simuliere LLM sagt nichts spezifisches -> Default FACT
|
||||
mock_llm.generate_raw_response.return_value = "Antwort."
|
||||
|
||||
mock_retriever = AsyncMock()
|
||||
mock_retriever.search.return_value.results = []
|
||||
|
||||
# Cache Reset
|
||||
import app.routers.chat
|
||||
app.routers.chat._DECISION_CONFIG_CACHE = None
|
||||
|
||||
req = ChatRequest(message="Hallo Welt", top_k=3)
|
||||
|
||||
response = await chat_endpoint(req, llm=mock_llm, retriever=mock_retriever)
|
||||
|
||||
# Prüfen ob System nicht crasht und vernünftige Defaults nutzt
|
||||
called_query_req = mock_retriever.search.call_args[0][0]
|
||||
|
||||
# FACT strategy hat in unserem Test Setup 'part_of': 2.0
|
||||
# Aber keine 'caused_by' boosts.
|
||||
self.assertEqual(response.intent, "FACT")
|
||||
self.assertNotIn("caused_by", called_query_req.boost_edges or {})
|
||||
|
||||
print("✅ Regression Test bestanden (Standard-Flow intakt).")
|
||||
|
||||
if __name__ == '__main__':
|
||||
# Async Support für Unittest
|
||||
loop = asyncio.new_event_loop()
|
||||
asyncio.set_event_loop(loop)
|
||||
unittest.main()
|
||||
Loading…
Reference in New Issue
Block a user