devils-agent / tests /test_retrieval.py
devildasdf's picture
Upload experimental BAIM code, research checkpoints and measured evaluations
795f737 verified
Raw
History Blame Contribute Delete
1.02 kB
import unittest
from baim.retrieval_audit import bm25
from baim.mind2web_smoke import normalize_task
class RetrievalTests(unittest.TestCase):
def test_broad_candidates_keep_tabs_without_allowing_sensitive_controls(self):
rows=[dict(role='tab',name='Account'),dict(role='generic',name='Account',sensitive=True),
dict(role='button',name='Other',enabled=False)]
self.assertEqual(bm25('Account',rows),[0])
def test_history_excludes_current_and_future_actions(self):
candidate=dict(tag='button',attributes='{}',backend_node_id='1')
step=dict(pos_candidates=[candidate],neg_candidates=[],cleaned_html='<button backend_node_id="1">A</button>',
action_uid='a',operation=dict(op='CLICK'))
task=dict(confirmed_task='Goal',action_reprs=['previous','current'],actions=[step,{**step,'action_uid':'b'}])
rows=list(normalize_task(task))
self.assertEqual(rows[0]['history'],[])
self.assertEqual(rows[1]['history'],['previous'])