73 lines
3.4 KiB
Python
73 lines
3.4 KiB
Python
import json
|
|
import sys
|
|
import unittest
|
|
from pathlib import Path
|
|
|
|
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
|
|
from pharmacy_agent import QUERY_SETTINGS, build_aliases, detect_drugs, prepare_query
|
|
from scripts.evaluation.run_evaluation import metrics, summarize
|
|
from scripts.evaluation.judge_answers import parse_verdict
|
|
|
|
|
|
def product(name, subs, rep="https://example.org/a"):
|
|
return dict(name=name,subs=subs,rep=rep,url=rep or "https://example.org/b",form="",rx="",type="",group="",ind="")
|
|
|
|
|
|
class AgentTests(unittest.TestCase):
|
|
def setUp(self):
|
|
self.products=[product("Nurofen 200 mg",["ibuprofén"]),product("Warfarin 5 mg",["warfarín"])]
|
|
self.aliases,self.prefixes,_=build_aliases(self.products)
|
|
|
|
def test_slovak_inflection_and_brand(self):
|
|
matches=detect_drugs("Nurofen spolu s warfarínom",self.aliases,self.prefixes)
|
|
self.assertEqual({m["label"].lower() for m in matches},{"nurofen","warfarín"})
|
|
|
|
def test_short_brand_and_full_name(self):
|
|
items=[product("ACC Long 600 mg",["acetylcysteín"])]
|
|
a,p,_=build_aliases(items)
|
|
self.assertEqual(len(detect_drugs("Na čo je ACC Long?",a,p)),1)
|
|
|
|
def test_ambiguous_brand_does_not_invent_single_substance(self):
|
|
items=[product("Brand 200 mg",["one"]),product("Brand Plus 200 mg",["one","two"])]
|
|
a,p,_=build_aliases(items)
|
|
result=prepare_query("Môžem užiť Brand?",items,a,p)
|
|
self.assertIn("nejednoznačný",result["user_prompt"])
|
|
self.assertNotIn("Brand = účinná látka one",result["query"])
|
|
|
|
def test_unrelated_prefix_not_detected(self):
|
|
self.assertEqual(detect_drugs("warfarinovyneznamyprodukt",self.aliases,self.prefixes),[])
|
|
|
|
def test_missing_data_is_not_no_interaction(self):
|
|
items=[product("Spofaplast náplasť",[],None)]
|
|
a,p,_=build_aliases(items)
|
|
result=prepare_query("Spofaplast a iné lieky?",items,a,p)
|
|
self.assertIn("neznamená",result["user_prompt"])
|
|
|
|
def test_source_identity_accepts_pil_and_detail(self):
|
|
case={"sources":["https://adc.sk/detail/x"]}
|
|
response={"references":[{"file_path":"https://adc.sk/pil/x","content":["text"]}]}
|
|
self.assertEqual(metrics(case,response,"mix")["source_hit"],1)
|
|
self.assertIsNone(metrics(case,response,"bypass")["source_hit"])
|
|
|
|
def test_failures_are_reported_not_scored_as_wrong(self):
|
|
result=summarize([{"mode":"mix","status":"error"},{"mode":"mix","status":"ok","seconds":2,"metrics":{"source_hit":1}}])
|
|
self.assertEqual(result["mix"]["failed"],1)
|
|
self.assertEqual(result["mix"]["source_hit"],1)
|
|
|
|
def test_graph_budget_reserves_original_source_text(self):
|
|
available=QUERY_SETTINGS["max_total_tokens"]-QUERY_SETTINGS["max_entity_tokens"]-QUERY_SETTINGS["max_relation_tokens"]
|
|
self.assertGreaterEqual(available,4000)
|
|
|
|
def test_judge_rejects_more_supported_claims_than_claims(self):
|
|
verdict={"grounding":{"total":2,"supported":3},"coverage":{"total":1,"supported":1},"appropriate_abstention":False}
|
|
with self.assertRaises(ValueError):
|
|
parse_verdict(json.dumps(verdict))
|
|
|
|
def test_judge_accepts_reasoning_prefix_without_scoring_it(self):
|
|
verdict={"grounding":{"total":2,"supported":1},"coverage":{"total":1,"supported":1},"appropriate_abstention":False}
|
|
self.assertEqual(parse_verdict("<think>not a verdict</think>"+json.dumps(verdict)),verdict)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|