def test_evaluate_fmeasure():
metrics = ['fmeasure']
ret_sql = f"""\
-- client: molehill/{molehill.__version__}
select
fmeasure(t.target, p.predicted) as fmeasure
from
prediction p
join
test t on (p.rowid = t.rowid)
;
"""
assert evaluate(metrics, 'target', 'predicted') == ret_sql
def test_evaluate_with_logloss():
metrics = ['logloss']
ret_sql = f"""\
-- client: molehill/{molehill.__version__}
select
logloss(p.probability, t.target) as logloss
from
prediction p
join
test t on (p.rowid = t.rowid)
;
"""
assert evaluate(metrics, 'target', 'predicted') == ret_sql
def test_evaluate_with_accuracy_precision_recall():
metrics = ['accuracy', 'precision', 'recall']
ret_sql = f"""\
-- client: molehill/{molehill.__version__}
select
cast(sum(if(p.predicted = t.target and t.target = 1, 1, 0)) as double)/count(1) as accuracy
, cast(sum(if(p.predicted = t.target and t.target = 1, 1, 0)) as double)/sum(if(p.predicted = 1, 1, 0)) as "precision"
, cast(sum(if(p.predicted = t.target and t.target = 1, 1, 0)) as double)/sum(if(t.target = 1, 1, 0)) as recall
from
prediction p
join
test t on (p.rowid = t.rowid)
;
"""
assert evaluate(metrics, 'target', 'predicted') == ret_sql
def test_evaluate_with_auc():
metrics = ['auc', 'logloss']
ret_sql = f"""\
-- client: molehill/{molehill.__version__}
select
auc(probability, target) as auc
, logloss(probability, target) as logloss
from
(
select
p.probability, t.target
from
pred p
join
actual t on (p.id = t.id)
order by
probability desc
) t2
;
"""
assert evaluate(metrics, 'target', 'probability', 'actual', 'pred',
'id') == ret_sql
def test_evaluate_unknown_measure():
metrics = ['unknown_metrics']
with pytest.raises(ValueError):
evaluate(metrics, 'target', 'predicted')