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

Enable and fix IVF PQ hypercube test #429

Draft
wants to merge 4 commits into
base: main
Choose a base branch
from
Draft

Enable and fix IVF PQ hypercube test #429

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2d8e8ec
enable and fix IVF PQ test
jparismorgan Jul 1, 2024
fbe2596
add debug lines, remove old debug lines
jparismorgan Jul 1, 2024
93e2999
test type
jparismorgan Jul 1, 2024
7d40bce
updates to debug ci test failure
jparismorgan Jul 23, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
8 changes: 4 additions & 4 deletions apis/python/test/conftest.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ def no_output(capfd):

# Fail if there is any output.
out, err = capfd.readouterr()
if out or err:
pytest.fail(
f"Test failed because output was captured. out:\n{out}\nerr:\n{err}"
)
# if out or err:
# pytest.fail(
# f"Test failed because output was captured. out:\n{out}\nerr:\n{err}"
# )
2 changes: 2 additions & 0 deletions src/include/index/ivf_pq_index.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1200,6 +1200,8 @@ class ivf_pq_index {
::num_vectors(*partitioned_pq_vectors_) == 0) {
read_index_infinite();
}
debug_partitioned_matrix(
*partitioned_pq_vectors_, "partitioned_pq_vectors_");
auto&& [active_partitions, active_queries] =
detail::ivf::partition_ivf_flat_index<indices_type>(
flat_ivf_centroids_, query_vectors, nprobe, num_threads_);
Expand Down
2 changes: 0 additions & 2 deletions src/include/test/unit_api_ivf_pq_index.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -860,8 +860,6 @@ TEST_CASE("write and load index with timestamps", "[api_ivf_pq_index]") {
scores_vector_array.num_vectors());
auto ids = std::span<uint32_t>(
(uint32_t*)ids_vector_array.data(), ids_vector_array.num_vectors());
debug_vector(scores, "scores");
debug_vector(ids, "ids");

CHECK(std::equal(
scores.begin(),
Expand Down
174 changes: 93 additions & 81 deletions src/include/test/unit_ivf_pq_index.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -373,16 +373,11 @@ TEST_CASE(
}
}

// Current code requires that the number of vectors in the training set be at
// least as large as the number of clusters.
//
#if 0
TEMPLATE_TEST_CASE(
"query stacked hypercube",
"[flativf_index]",
float,
uint8_t) {
size_t k_dist = GENERATE(0, 32);
TEMPLATE_TEST_CASE("query stacked hypercube", "[flativf_index]", float) {
// size_t k_dist = GENERATE(0, 32);
size_t k_dist = 32;
std::cout << "k_dist: " << k_dist
<< " -----------------------------------------" << std::endl;
size_t k_near = k_dist;
size_t k_far = k_dist;

Expand All @@ -406,74 +401,96 @@ TEMPLATE_TEST_CASE(
hypercube4(j + 9, i) = hypercube1(j, i);
}
}
SECTION("nlist = 1") {
size_t k_nn = 6;
size_t nlist = 1;

auto ivf_idx2 = ivf_pq_index<TestType, uint32_t, uint32_t>(
/*128,*/ nlist, 2, 4, 1.e-4); // dim nlist maxiter eps nthreads
ivf_idx2.train_ivf(hypercube2);
ivf_idx2.add(hypercube2, ids);
auto ivf_idx4 = ivf_pq_index<TestType, uint32_t, uint32_t>(
/*128,*/ nlist, 2, 4, 1.e-4);
ivf_idx4.train_ivf(hypercube4);
ivf_idx4.add(hypercube4, ids);

auto top_k_ivf_scores = ColMajorMatrix<float>();
auto top_k_ivf = ColMajorMatrix<unsigned>();
auto top_k_scores = ColMajorMatrix<float>();
auto top_k = ColMajorMatrix<uint64_t>();
auto query2 = ColMajorMatrix<TestType>();
auto query4 = ColMajorMatrix<TestType>();

SECTION("query2/4 = 0...") {
query2 = ColMajorMatrix<TestType>{{0, 0, 0, 0, 0, 0}};
query4 = ColMajorMatrix<TestType>{{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}};
}
SECTION("query2/4 = 127...") {
query2 = ColMajorMatrix<TestType>{{127, 127, 127, 127, 127, 127}};
query4 = ColMajorMatrix<TestType>{
{127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127}};
}
SECTION("query2/4 = 0...") {
query2 = ColMajorMatrix<TestType>{{0, 0, 0, 127, 127, 127}};
query4 = ColMajorMatrix<TestType>{
{0, 0, 0, 0, 0, 0, 127, 127, 127, 127, 127, 127}};
}
SECTION("query2/4 = 127...") {
query2 = ColMajorMatrix<TestType>{{127, 127, 127, 0, 0, 0}};
query4 = ColMajorMatrix<TestType>{
{127, 127, 127, 127, 127, 127, 0, 0, 0, 0, 0, 0}};
}
SECTION("query2/4 = 127...") {
query2 = ColMajorMatrix<TestType>{
{127, 0, 127, 0, 127, 0}, {0, 127, 0, 127, 0, 127}};
query4 = ColMajorMatrix<TestType>{
{127, 0, 127, 0, 127, 0, 127, 0, 127, 0, 127, 0},
{0, 127, 0, 127, 0, 127, 0, 127, 0, 127, 0, 127}};
}
debug_matrix(hypercube0, "hypercube0");
debug_matrix(hypercube1, "hypercube1");
debug_matrix(hypercube2, "hypercube2");
debug_matrix(hypercube4, "hypercube4");
debug_vector(ids, "ids");

std::tie(top_k_scores, top_k) = detail::flat::qv_query_heap(
hypercube2, query2, k_nn, 1, sum_of_squares_distance{});
std::tie(top_k_ivf_scores, top_k_ivf) =
ivf_idx2.query_infinite_ram(query2, k_nn, 1); // k, nprobe
size_t intersections0 = count_intersections(top_k_ivf, top_k, k_nn);
double recall0 = intersections0 / ((double)top_k.num_cols() * k_nn);
CHECK(intersections0 == k_nn * num_vectors(query2));
CHECK(recall0 == 1.0);

std::tie(top_k_scores, top_k) = detail::flat::qv_query_heap(
hypercube4, query4, k_nn, 1, sum_of_squares_distance{});
std::tie(top_k_ivf_scores, top_k_ivf) =
ivf_idx4.query_infinite_ram(query4, k_nn, 1); // k, nprobe

size_t intersections1 = (long)count_intersections(top_k_ivf, top_k, k_nn);
double recall1 = intersections1 / ((double)top_k.num_cols() * k_nn);
CHECK(intersections1 == k_nn * num_vectors(query4));
CHECK(recall1 == 1.0);
}
size_t k_nn = 6;
size_t nlist = 1;
size_t num_subspaces = 3;
size_t max_iter = 4;
float tol = 1.e-4;

auto ivf_idx2 = ivf_pq_index<TestType, uint32_t, uint32_t>(
nlist, dimensions(hypercube2), max_iter, tol);
ivf_idx2.train_ivf(hypercube2);
ivf_idx2.add(hypercube2, ids);
auto ivf_idx4 = ivf_pq_index<TestType, uint32_t, uint32_t>(
nlist, dimensions(hypercube4), max_iter, tol);
ivf_idx4.train_ivf(hypercube4);
ivf_idx4.add(hypercube4, ids);

auto query2 = ColMajorMatrix<TestType>();
auto query4 = ColMajorMatrix<TestType>();
// SECTION("query with all 0's") {
// std::cout << "query with all 0's ------------" << std::endl;
// query2 = ColMajorMatrix<TestType>{{0, 0, 0, 0, 0, 0}};
// query4 = ColMajorMatrix<TestType>{{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}};
// }
// SECTION("query with all 127's") {
// std::cout << "query with all 127's ------------" << std::endl;
// query2 = ColMajorMatrix<TestType>{{127, 127, 127, 127, 127, 127}};
// query4 = ColMajorMatrix<TestType>{
// {127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127}};
// }
// SECTION("query with all 0's and then 127's") {
// std::cout << "query with all 0's and then 127's ------------" <<
// std::endl; query2 = ColMajorMatrix<TestType>{{0, 0, 0, 127, 127, 127}};
// query4 = ColMajorMatrix<TestType>{
// {0, 0, 0, 0, 0, 0, 127, 127, 127, 127, 127, 127}};
// }
// SECTION("query with all 127's and then 0's") {
// std::cout << "query with all 127's and then 0's ------------" <<
// std::endl; query2 = ColMajorMatrix<TestType>{{127, 127, 127, 0, 0, 0}};
// query4 = ColMajorMatrix<TestType>{
// {127, 127, 127, 127, 127, 127, 0, 0, 0, 0, 0, 0}};
// }
SECTION("query with alternating 127's and 0's") {
std::cout << "query with alternating 127's and 0's ------------"
<< std::endl;
query2 = ColMajorMatrix<TestType>{
{127, 0, 127, 0, 127, 0}, {0, 127, 0, 127, 0, 127}};
query4 = ColMajorMatrix<TestType>{
{127, 0, 127, 0, 127, 0, 127, 0, 127, 0, 127, 0},
{0, 127, 0, 127, 0, 127, 0, 127, 0, 127, 0, 127}};
}

auto top_k_scores = ColMajorMatrix<float>();
auto top_k_ids = ColMajorMatrix<uint64_t>();
auto top_k_ivf_scores = ColMajorMatrix<float>();
auto top_k_ivf_ids = ColMajorMatrix<uint32_t>();

std::cout << "query2 ~~~~~~" << std::endl;
std::tie(top_k_scores, top_k_ids) = detail::flat::qv_query_heap(
hypercube2, query2, k_nn, 1, sum_of_squares_distance{});
debug_matrix(top_k_ids, "top_k_ids");
debug_matrix(top_k_scores, "top_k_scores");
std::tie(top_k_ivf_scores, top_k_ivf_ids) =
ivf_idx2.query_infinite_ram(query2, k_nn, nlist);
debug_matrix(top_k_ivf_ids, "top_k_ivf_ids");
debug_matrix(top_k_ivf_scores, "top_k_ivf_scores");
size_t intersections0 = count_intersections(top_k_ivf_ids, top_k_ids, k_nn);
double recall0 = intersections0 / ((double)top_k_ids.num_cols() * k_nn);
CHECK(intersections0 == k_nn * num_vectors(query2));
CHECK(recall0 == 1.0);

std::cout << "query4 ~~~~~~" << std::endl;
std::tie(top_k_scores, top_k_ids) = detail::flat::qv_query_heap(
hypercube4, query4, k_nn, 1, sum_of_squares_distance{});
debug_matrix(top_k_ids, "top_k_ids");
debug_matrix(top_k_scores, "top_k_scores");
std::tie(top_k_ivf_scores, top_k_ivf_ids) =
ivf_idx4.query_infinite_ram(query4, k_nn, nlist);
debug_matrix(top_k_ivf_ids, "top_k_ivf_ids");
debug_matrix(top_k_ivf_scores, "top_k_ivf_scores");
size_t intersections1 =
(long)count_intersections(top_k_ivf_ids, top_k_ids, k_nn);
double recall1 = intersections1 / ((double)top_k_ids.num_cols() * k_nn);
CHECK(intersections1 == k_nn * num_vectors(query4));
CHECK(recall1 == 1.0);
}
#endif

TEST_CASE("Build index and query in place, infinite", "[ivf_pq_index]") {
tiledb::Context ctx;
Expand Down Expand Up @@ -592,7 +609,6 @@ TEST_CASE("query empty index", "[ivf_pq_index]") {
{
auto data =
ColMajorMatrixWithIds<siftsmall_feature_type>(dimensions, num_vectors);
debug_matrix_with_ids(data, "data");
index.train(data, data.raveled_ids());
index.add(data, data.raveled_ids());
}
Expand Down Expand Up @@ -667,8 +683,6 @@ TEST_CASE("query simple", "[ivf_pq_index]") {
auto value = static_cast<feature_type>(i);
auto queries = ColMajorMatrix<feature_type>{{value, value, value, value}};
auto&& [scores, ids] = index.query_infinite_ram(queries, k_nn, nprobe);
debug_matrix(scores, "scores");
debug_matrix(ids, "ids");
CHECK(scores(0, 0) == 0);
CHECK(ids(0, 0) == i * 11);
}
Expand All @@ -688,8 +702,6 @@ TEST_CASE("query simple", "[ivf_pq_index]") {
auto value = static_cast<feature_type>(i);
auto queries = ColMajorMatrix<feature_type>{{value, value, value, value}};
auto&& [scores, ids] = index.query_infinite_ram(queries, k_nn, nprobe);
debug_matrix(scores, "scores");
debug_matrix(ids, "ids");
CHECK(scores(0, 0) == 0);
CHECK(ids(0, 0) == i * 11);
}
Expand Down