def test_check_scores(session_scope_module):
# check both set_scores and get_scores
submission_id = 1
path_results = os.path.join(HERE, 'data', 'iris_predictions')
set_scores(session_scope_module, submission_id, path_results)
scores = get_scores(session_scope_module, submission_id)
multi_index = pd.MultiIndex.from_product(
[[0, 1], ['train', 'valid', 'test']], names=['fold', 'step']
)
expected_df = pd.DataFrame(
{'acc': [0.604167, 0.583333, 0.733333, 0.604167, 0.583333, 0.733333],
'error': [0.395833, 0.416667, 0.266667, 0.395833, 0.416667, 0.266667],
'nll': [0.732763, 2.194549, 0.693464, 0.746132, 2.030762, 0.693992],
'f1_70': [0.333333, 0.33333, 0.666667, 0.33333, 0.33333, 0.666667]},
index=multi_index
)
assert_frame_equal(scores, expected_df, check_less_precise=True)
def test_score_submission(session_scope_module):
submission_id = 9
multi_index = pd.MultiIndex.from_product(
[[0, 1], ['train', 'valid', 'test']], names=['fold', 'step']
)
expected_df = pd.DataFrame(
{'acc': [0.604167, 0.583333, 0.733333, 0.604167, 0.583333, 0.733333],
'error': [0.395833, 0.416667, 0.266667, 0.395833, 0.416667, 0.266667],
'nll': [0.732763, 2.194549, 0.693464, 0.746132, 2.030762, 0.693992],
'f1_70': [0.333333, 0.33333, 0.666667, 0.33333, 0.33333, 0.666667]},
index=multi_index
)
path_results = os.path.join(HERE, 'data', 'iris_predictions')
with pytest.raises(ValueError, match='Submission state must be "tested"'):
score_submission(session_scope_module, submission_id)
set_submission_state(session_scope_module, submission_id, 'tested')
set_predictions(session_scope_module, submission_id, path_results)
score_submission(session_scope_module, submission_id)
scores = get_scores(session_scope_module, submission_id)
assert_frame_equal(scores, expected_df, check_less_precise=True)