def test_recall_array():
prec = lm.recall(np.array([1, 3]), np.array([1, 3]))
assert prec == approx(1.0)
prec = lm.recall(np.array([1, 2, 3, 4]), np.array([1, 3, 5, 7]))
assert prec == approx(0.5)
prec = lm.recall(np.array([1, 2, 3, 4]), np.arange(4, 9, 1, 'u4'))
assert prec == approx(0.2)
def test_recall_series_index():
prec = lm.recall(pd.Series([1, 3]), pd.Index([1, 3]))
assert prec == approx(1.0)
prec = lm.recall(pd.Series([1, 2, 3, 4]), pd.Index([1, 3, 5, 7]))
assert prec == approx(0.5)
prec = lm.recall(pd.Series([1, 2, 3, 4]), pd.Index(range(4, 9)))
assert prec == approx(0.2)
def test_recall_bulk_k(demo_recs):
"bulk and normal match"
train, test, recs = demo_recs
assert test['user'].value_counts().max() > 5
rla = topn.RecListAnalysis()
rla.add_metric(recall, name='rk', k=5)
rla.add_metric(recall)
# metric without the bulk capabilities
rla.add_metric(lambda *a, **k: recall(*a, **k), name='ind_rk', k=5)
rla.add_metric(lambda *a: recall(*a), name='ind_r')
res = rla.compute(recs, test)
print(res)
_log.info('recall mismatches:\n%s', res[res.recall != res.ind_r])
assert res.recall.values == approx(res.ind_r.values)
assert res.rk.values == approx(res.ind_rk.values)
def test_pr_bulk_match(demo_recs, drop_rating):
"bulk and normal match"
train, test, recs = demo_recs
if drop_rating:
test = test[['user', 'item']]
rla = topn.RecListAnalysis()
rla.add_metric(precision)
rla.add_metric(recall)
# metric without the bulk capabilities
rla.add_metric(lambda *a: precision(*a), name='ind_p')
rla.add_metric(lambda *a: recall(*a), name='ind_r')
res = rla.compute(recs, test)
print(res)
_log.info('precision mismatches:\n%s', res[res.precision != res.ind_p])
_log.info('recall mismatches:\n%s', res[res.recall != res.ind_r])
assert res.precision.values == approx(res.ind_p.values)
assert res.recall.values == approx(res.ind_r.values)
def test_recall_simple_cases():
prec = lm.recall([1, 3], [1, 3])
assert prec == approx(1.0)
prec = lm.recall([1], [1, 3])
assert prec == approx(0.5)
prec = lm.recall([1, 2, 3, 4], [1, 3])
assert prec == approx(1.0)
prec = lm.recall([1, 2, 3, 4], [1, 3, 5])
assert prec == approx(2.0 / 3)
prec = lm.recall([1, 2, 3, 4], range(5, 10))
assert prec == approx(0.0)
prec = lm.recall([1, 2, 3, 4], range(4, 9))
assert prec == approx(0.2)
def _test_recall(items, rel, **kwargs):
recs = pd.DataFrame({'item': items})
truth = pd.DataFrame({'item': rel}).set_index('item')
return recall(recs, truth, **kwargs)
def test_recall_empty_zero():
prec = lm.recall([], [1, 3])
assert prec == approx(0)
def test_recall_series_set():
prec = lm.recall(pd.Series([1, 2, 3, 4]), [1, 3, 5, 7])
assert prec == approx(0.5)
prec = lm.recall(pd.Series([1, 2, 3, 4]), range(4, 9))
assert prec == approx(0.2)
def test_recall_norel_na():
prec = lm.recall([1, 3], [])
assert np.isnan(prec)