Files
milvus/internal/core/src/query/SearchBruteForceSparseTest.cpp
T
Spade AandGitHub f6f716bcfd feat: impl StructArray -- support embedding searches embeddings in embedding list with element level filter expression (#45830)
issue: https://github.com/milvus-io/milvus/issues/42148

For a vector field inside a STRUCT, since a STRUCT can only appear as
the element type of an ARRAY field, the vector field in STRUCT is
effectively an array of vectors, i.e. an embedding list.
Milvus already supports searching embedding lists with metrics whose
names start with the prefix MAX_SIM_.

This PR allows Milvus to search embeddings inside an embedding list
using the same metrics as normal embedding fields. Each embedding in the
list is treated as an independent vector and participates in ANN search.

Further, since STRUCT may contain scalar fields that are highly related
to the embedding field, this PR introduces an element-level filter
expression to refine search results.
The grammar of the element-level filter is:

element_filter(structFieldName, $[subFieldName] == 3)

where $[subFieldName] refers to the value of subFieldName in each
element of the STRUCT array structFieldName.

It can be combined with existing filter expressions, for example:

"varcharField == 'aaa' && element_filter(struct_field, $[struct_int] ==
3)"

A full example:
```
struct_schema = milvus_client.create_struct_field_schema()
struct_schema.add_field("struct_str", DataType.VARCHAR, max_length=65535)
struct_schema.add_field("struct_int", DataType.INT32)
struct_schema.add_field("struct_float_vec", DataType.FLOAT_VECTOR, dim=EMBEDDING_DIM)

schema.add_field(
    "struct_field",
    datatype=DataType.ARRAY,
    element_type=DataType.STRUCT,
    struct_schema=struct_schema,
    max_capacity=1000,
)
...

filter = "varcharField == 'aaa' && element_filter(struct_field, $[struct_int] == 3 && $[struct_str] == 'abc')"
res = milvus_client.search(
    COLLECTION_NAME,
    data=query_embeddings,
    limit=10,
    anns_field="struct_field[struct_float_vec]",
    filter=filter,
    output_fields=["struct_field[struct_int]", "varcharField"],
)

```
TODO:
1. When an `element_filter` expression is used, a regular filter
expression must also be present. Remove this restriction.
2. Implement `element_filter` expressions in the `query`.

---------

Signed-off-by: SpadeA <tangchenjie1210@gmail.com>
2025-12-15 12:01:15 +08:00

186 lines
6.5 KiB
C++

// Copyright (C) 2019-2024 Zilliz. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software distributed under the License
// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
// or implied. See the License for the specific language governing permissions and limitations under the License
#include <gtest/gtest.h>
#include <map>
#include <random>
#include "common/Utils.h"
#include "query/SearchBruteForce.h"
#include "test_utils/Constants.h"
#include "test_utils/Distance.h"
#include "test_utils/DataGen.h"
using namespace milvus;
using namespace milvus::segcore;
using namespace milvus::query;
namespace {
std::vector<int>
SearchRef(const knowhere::sparse::SparseRow<milvus::SparseValueType>* base,
const knowhere::sparse::SparseRow<milvus::SparseValueType>& query,
int nb,
int topk) {
std::vector<std::tuple<float, int>> res;
for (int i = 0; i < nb; i++) {
auto& row = base[i];
auto distance = row.dot(query);
res.emplace_back(-distance, i);
}
std::sort(res.begin(), res.end());
std::vector<int> offsets;
for (int i = 0; i < topk; i++) {
auto [distance, offset] = res[i];
if (distance == 0) {
distance = std::numeric_limits<float>::quiet_NaN();
offset = -1;
}
offsets.push_back(offset);
}
return offsets;
}
std::vector<int>
RangeSearchRef(
const knowhere::sparse::SparseRow<milvus::SparseValueType>* base,
const knowhere::sparse::SparseRow<milvus::SparseValueType>& query,
int nb,
float radius,
float range_filter,
int topk) {
std::vector<int> offsets;
for (int i = 0; i < nb; i++) {
auto& row = base[i];
auto distance = row.dot(query);
if (distance <= range_filter && distance > radius) {
offsets.push_back(i);
}
}
// select and sort top k on the range filter side
std::sort(offsets.begin(), offsets.end(), [&](int a, int b) {
return base[a].dot(query) > base[b].dot(query);
});
if (offsets.size() > topk) {
offsets.resize(topk);
}
return offsets;
}
void
AssertMatch(const std::vector<int>& expected, const int64_t* actual) {
for (int i = 0; i < expected.size(); i++) {
ASSERT_EQ(expected[i], actual[i]);
}
}
bool
is_supported_sparse_float_metric(const std::string& metric) {
return milvus::IsMetricType(metric, knowhere::metric::IP);
}
} // namespace
class TestSparseFloatSearchBruteForce : public ::testing::Test {
public:
void
Run(int nb, int nq, int topk, const knowhere::MetricType& metric_type) {
auto bitset = std::make_shared<BitsetType>();
bitset->resize(nb);
auto bitset_view = BitsetView(*bitset);
auto base = milvus::segcore::GenerateRandomSparseFloatVector(nb);
auto query = milvus::segcore::GenerateRandomSparseFloatVector(nq);
auto index_info = std::map<std::string, std::string>{};
SearchInfo search_info;
search_info.topk_ = topk;
search_info.metric_type_ = metric_type;
dataset::SearchDataset query_dataset{
metric_type, nq, topk, -1, kTestSparseDim, query.get()};
auto raw_dataset =
query::dataset::RawDataset{0, kTestSparseDim, nb, base.get()};
if (!is_supported_sparse_float_metric(metric_type)) {
ASSERT_ANY_THROW(BruteForceSearch(query_dataset,
raw_dataset,
search_info,
index_info,
bitset_view,
DataType::VECTOR_SPARSE_U32_F32,
DataType::NONE,
nullptr));
return;
}
auto result = BruteForceSearch(query_dataset,
raw_dataset,
search_info,
index_info,
bitset_view,
DataType::VECTOR_SPARSE_U32_F32,
DataType::NONE,
nullptr);
for (int i = 0; i < nq; i++) {
auto ref = SearchRef(base.get(), *(query.get() + i), nb, topk);
auto ans = result.get_offsets() + i * topk;
AssertMatch(ref, ans);
}
search_info.search_params_[RADIUS] = 0.1;
search_info.search_params_[RANGE_FILTER] = 0.5;
auto result2 = BruteForceSearch(query_dataset,
raw_dataset,
search_info,
index_info,
bitset_view,
DataType::VECTOR_SPARSE_U32_F32,
DataType::NONE,
nullptr);
for (int i = 0; i < nq; i++) {
auto ref = RangeSearchRef(
base.get(), *(query.get() + i), nb, 0.1, 0.5, topk);
auto ans = result2.get_offsets() + i * topk;
AssertMatch(ref, ans);
}
auto result3 = PackBruteForceSearchIteratorsIntoSubResult(
query_dataset,
raw_dataset,
search_info,
index_info,
bitset_view,
DataType::VECTOR_SPARSE_U32_F32);
auto iterators = result3.chunk_iterators();
for (int i = 0; i < nq; i++) {
auto it = iterators[i];
auto q = *(query.get() + i);
auto last_dis = std::numeric_limits<float>::max();
// we should see strict decreasing distances for brute force iterator.
while (it->HasNext()) {
auto [offset, dis] = it->Next();
ASSERT_LE(dis, last_dis);
last_dis = dis;
ASSERT_FLOAT_EQ(dis, base[offset].dot(q));
}
}
}
};
TEST_F(TestSparseFloatSearchBruteForce, NotSupported) {
Run(100, 10, 5, "L2");
Run(100, 10, 5, "l2");
Run(100, 10, 5, "lxxx");
}
TEST_F(TestSparseFloatSearchBruteForce, IP) {
Run(100, 10, 5, "IP");
Run(100, 10, 5, "ip");
}