Skip to content

fix: calculate relevant docs on index instead of queries #950

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
samsja merged 5 commits into from
Dec 16, 2022
Merged

fix: calculate relevant docs on index instead of queries #950

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6f2f5b6
fix: calculate relevant docs on index instead of queries
guenthermi Dec 15, 2022
17c3f9e
refactor: reduce num labels if necessary
guenthermi Dec 15, 2022
7900014
test: add test for exclude_self
guenthermi Dec 15, 2022
e1a74b9
refactor: change metrics in tests
guenthermi Dec 15, 2022
017354d
chore: merge main into branch
guenthermi Dec 16, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 8 additions & 6 deletions docarray/array/mixins/evaluation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -458,14 +458,16 @@ def fuse_matches(global_matches: DocumentArray, local_matches: DocumentArray):
new_matches.append(m)
query_data[doc.id, 'matches'] = new_matches

if ground_truth and label_tag in ground_truth[0].tags:
num_relevant_documents_per_label = dict(
Counter([d.tags[label_tag] for d in ground_truth])
)
elif not ground_truth and label_tag in query_data[0].tags:
if (
not (ground_truth and len(ground_truth) > 0 and ground_truth[0].matches)
and label_tag in index_data[0].tags
):
num_relevant_documents_per_label = dict(
Counter([d.tags[label_tag] for d in query_data])
Counter([d.tags[label_tag] for d in index_data])
)
if only_one_dataset and exclude_self:
for k, v in num_relevant_documents_per_label.items():
num_relevant_documents_per_label[k] -= 1
else:
num_relevant_documents_per_label = None

Expand Down
94 changes: 92 additions & 2 deletions tests/unit/array/mixins/oldproto/test_eval_class.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -528,6 +528,52 @@ def test_embed_and_evaluate_single_da(storage, config, start_storage):
assert all([v == 1.0 for v in res.values()])


@pytest.mark.parametrize(
'exclude_self, expected_results',
[
(
True,
{
'r_precision': 1,
'precision_at_k': 1.0,
'hit_at_k': 1.0,
'average_precision': 1.0,
'reciprocal_rank': 1.0,
'recall_at_k': 5.0 / 9.0,
'f1_score_at_k': (10.0 / 9.0) / (5.0 / 9.0 + 1),
},
),
(
False,
{
'r_precision': 1.0,
'precision_at_k': 1.0,
'hit_at_k': 1.0,
'average_precision': 1.0,
'reciprocal_rank': 1.0,
'recall_at_k': 0.5,
'f1_score_at_k': 1.0 / 1.5,
},
),
],
)
def test_embed_and_evaluate_with_and_without_exclude_self(
exclude_self, expected_results
):
queries_da = DocumentArray(
[Document(text=str(i % 10), label=i % 10) for i in range(100)]
)
res = queries_da.embed_and_evaluate(
metrics=list(expected_results.keys()),
exclude_self=exclude_self,
embed_funcs=dummy_embed_function,
match_batch_size=1,
limit=5,
)
for key in expected_results:
assert abs(res[key] - expected_results[key]) < 1e-5


@pytest.mark.parametrize(
'sample_size',
[None, 10],
Expand Down Expand Up @@ -567,6 +613,35 @@ def test_embed_and_evaluate_two_das(storage, config, sample_size, start_storage)
assert all([v == 1.0 for v in res.values()])


def test_embed_and_evaluate_two_different_das():
queries_da = DocumentArray([Document(text=str(i), label=i % 10) for i in range(10)])
index_da = DocumentArray(
[Document(text=str(i % 10), label=i % 10) for i in range(10, 210)]
)
res = queries_da.embed_and_evaluate(
index_data=index_da,
metrics=[
'r_precision',
'precision_at_k',
'hit_at_k',
'average_precision',
'reciprocal_rank',
'recall_at_k',
'f1_score_at_k',
],
embed_funcs=dummy_embed_function,
match_batch_size=1,
limit=10,
)
assert res['r_precision'] == 1.0
assert res['precision_at_k'] == 1.0
assert res['hit_at_k'] == 1.0
assert res['average_precision'] == 1.0
assert res['reciprocal_rank'] == 1.0
assert res['recall_at_k'] == 0.5
assert abs(res['f1_score_at_k'] - 1.0 / 1.5) < 1e-5


@pytest.mark.parametrize(
'use_index, expected, label_tag',
[
Expand Down Expand Up @@ -639,7 +714,16 @@ def emb_func(da):
],
)
def test_embed_and_evaluate_on_real_data(two_embed_funcs, kwargs):
metric_names = ['precision_at_k', 'reciprocal_rank']
metric_names = [
'r_precision',
'precision_at_k',
'hit_at_k',
'average_precision',
'reciprocal_rank',
'recall_at_k',
'f1_score_at_k',
'dcg_at_k',
]

labels = ['18828_alt.atheism', '18828_comp.graphics']
news = [load_dataset('newsgroup', label) for label in labels]
Expand Down Expand Up @@ -673,10 +757,16 @@ def emb_func(da):
)

# re-calculate manually
num_relevant_documents_per_label = dict(
Counter([d.tags['label'] for d in index_docs])
)
emb_func(query_docs)
emb_func(index_docs)
query_docs.match(index_docs)
res2 = query_docs.evaluate(metrics=metric_names)
res2 = query_docs.evaluate(
metrics=metric_names,
num_relevant_documents_per_label=num_relevant_documents_per_label,
)

for key in res:
assert key in res2
Expand Down