Skip to content
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

Jump to bottom

InnerProduct testing for CAGRA+HNSW #2297

Merged
merged 6 commits into from
May 23, 2024
Merged

InnerProduct testing for CAGRA+HNSW #2297

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
2c748b6
add ip tests
divyegala May 7, 2024
ff19708
format
divyegala May 8, 2024
ed370d6
style check
divyegala May 8, 2024
c6b10b0
line too long
divyegala May 8, 2024
1d7c6e5
Merge branch 'branch-24.06' into ip-cagra-hnsw
cjnolet May 15, 2024
7915ed5
Merge branch 'branch-24.06' into ip-cagra-hnsw
divyegala May 21, 2024
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
27 changes: 24 additions & 3 deletions python/pylibraft/pylibraft/test/test_hnsw.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,18 +29,26 @@ def run_hnsw_build_search_test(
k=10,
dtype=np.float32,
metric="sqeuclidean",
build_algo="ivf_pq",
intermediate_graph_degree=128,
graph_degree=64,
search_params={},
):
dataset = generate_data((n_rows, n_cols), dtype)
if metric == "inner_product":
dataset = normalize(dataset, norm="l2", axis=1)
if dtype in [np.int8, np.uint8]:
pytest.skip(
"inner_product metric is not supported for int8/uint8 data"
)
if build_algo == "nn_descent":
pytest.skip("inner_product metric is not supported for nn_descent")

build_params = cagra.IndexParams(
metric=metric,
intermediate_graph_degree=intermediate_graph_degree,
graph_degree=graph_degree,
build_algo=build_algo,
)

index = cagra.build(build_params, dataset)
Expand All @@ -57,7 +65,14 @@ def run_hnsw_build_search_test(
out_dist, out_idx = hnsw.search(search_params, hnsw_index, queries, k)

# Calculate reference values with sklearn
nn_skl = NearestNeighbors(n_neighbors=k, algorithm="brute", metric=metric)
skl_metric = {
"sqeuclidean": "sqeuclidean",
"inner_product": "cosine",
"euclidean": "euclidean",
}[metric]
nn_skl = NearestNeighbors(
n_neighbors=k, algorithm="brute", metric=skl_metric
)
nn_skl.fit(dataset)
skl_idx = nn_skl.kneighbors(queries, return_distance=False)

Expand All @@ -69,9 +84,15 @@ def run_hnsw_build_search_test(
@pytest.mark.parametrize("k", [10, 20])
@pytest.mark.parametrize("ef", [30, 40])
@pytest.mark.parametrize("num_threads", [2, 4])
def test_hnsw(dtype, k, ef, num_threads):
@pytest.mark.parametrize("metric", ["sqeuclidean", "inner_product"])
@pytest.mark.parametrize("build_algo", ["ivf_pq", "nn_descent"])
def test_hnsw(dtype, k, ef, num_threads, metric, build_algo):
# Note that inner_product tests use normalized input which we cannot
# represent in int8, therefore we test only sqeuclidean metric here.
run_hnsw_build_search_test(
dtype=dtype, k=k, search_params={"ef": ef, "num_threads": num_threads}
dtype=dtype,
k=k,
metric=metric,
build_algo=build_algo,
search_params={"ef": ef, "num_threads": num_threads},
)
Loading