postgres/typesense
1
0
Fork 0
mirror of https://github.com/typesense/typesense.git synced 2025-05-23 23:30:42 +08:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Merge pull request #1083 from ozanarmagan/v0.25-join

Browse Source 
Improve HybridSearchWithExplicitVector test
This commit is contained in:
Kishore Nallan 2023-07-03 16:09:12 +05:30 committed by GitHub
commit 6692e87d73
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 53 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

75
test/collection_vector_search_test.cpp
View File

@ -679,42 +679,73 @@ TEST_F(CollectionVectorTest, VectorWithNullValue) {
}
TEST_F(CollectionVectorTest, HybridSearchWithExplicitVector) {
nlohmann::json schema = R"({
"name": "coll1",
"fields": [
{"name": "name", "type": "string"},
{"name": "vec", "type": "float[]", "embed":{"from": ["name"], "model_config": {"model_name": "ts/e5-small"}}}
]
})"_json;
nlohmann::json schema = R"({
"name": "objects",
"fields": [
{"name": "name", "type": "string"},
{"name": "embedding", "type":"float[]", "embed":{"from": ["name"], "model_config": {"model_name": "ts/e5-small"}}}
]
})"_json;
TextEmbedderManager::set_model_dir("/tmp/typesense_test/models");
Collection* coll1 = collectionManager.create_collection(schema).get();
auto op = collectionManager.create_collection(schema);
ASSERT_TRUE(op.ok());
Collection* coll = op.get();
nlohmann::json object;
object["name"] = "butter";
auto add_op = coll->add(object.dump());
ASSERT_TRUE(add_op.ok());
nlohmann::json doc;
object["name"] = "butterball";
add_op = coll->add(object.dump());
ASSERT_TRUE(add_op.ok());
doc["name"] = "john doe";
ASSERT_TRUE(coll1->add(doc.dump()).ok());
object["name"] = "butterfly";
add_op = coll->add(object.dump());
ASSERT_TRUE(add_op.ok());
std::string dummy_vec_string = "[0.9";
for (int i = 0; i < 382; i++) {
dummy_vec_string += ", 0.9";
nlohmann::json model_config = R"({
"model_name": "ts/e5-small"
})"_json;
auto query_embedding = TextEmbedderManager::get_instance().get_text_embedder(model_config).get()->Embed("butter");
std::string vec_string = "[";
for(size_t i = 0; i < query_embedding.embedding.size(); i++) {
vec_string += std::to_string(query_embedding.embedding[i]);
if(i != query_embedding.embedding.size() - 1) {
vec_string += ",";
}
}
dummy_vec_string += ", 0.9]";
auto results_op = coll1->search("john", {"name"}, "", {}, {}, {0}, 20, 1, FREQUENCY, {true}, Index::DROP_TOKENS_THRESHOLD,
vec_string += "]";
auto search_res_op = coll->search("butter", {"name"}, "", {}, {}, {0}, 20, 1, FREQUENCY, {true}, Index::DROP_TOKENS_THRESHOLD,
spp::sparse_hash_set<std::string>(),
spp::sparse_hash_set<std::string>(), 10, "", 30, 5,
"", 10, {}, {}, {}, 0,
"<mark>", "</mark>", {}, 1000, true, false, true, "", false, 6000 * 1000, 4, 7,
fallback,
4, {off}, 32767, 32767, 2,
false, true, "vec:(" + dummy_vec_string +")");
ASSERT_EQ(true, results_op.ok());
ASSERT_EQ(1, results_op.get()["found"].get<size_t>());
ASSERT_EQ(1, results_op.get()["hits"].size());
false, true, "embedding:(" + vec_string + ")");
ASSERT_TRUE(search_res_op.ok());
auto search_res = search_res_op.get();
ASSERT_EQ(3, search_res["found"].get<size_t>());
ASSERT_EQ(3, search_res["hits"].size());
// Hybrid search with rank fusion order:
// 1. butter (1/1 * 0.7) + (1/1 * 0.3) = 1
// 2. butterfly (1/2 * 0.7) + (1/3 * 0.3) = 0.45
// 3. butterball (1/3 * 0.7) + (1/2 * 0.3) = 0.383
ASSERT_EQ("butter", search_res["hits"][0]["document"]["name"].get<std::string>());
ASSERT_EQ("butterfly", search_res["hits"][1]["document"]["name"].get<std::string>());
ASSERT_EQ("butterball", search_res["hits"][2]["document"]["name"].get<std::string>());
ASSERT_FLOAT_EQ((1.0/1.0 * 0.7) + (1.0/1.0 * 0.3), search_res["hits"][0]["hybrid_search_info"]["rank_fusion_score"].get<float>());
ASSERT_FLOAT_EQ((1.0/2.0 * 0.7) + (1.0/3.0 * 0.3), search_res["hits"][1]["hybrid_search_info"]["rank_fusion_score"].get<float>());
ASSERT_FLOAT_EQ((1.0/3.0 * 0.7) + (1.0/2.0 * 0.3), search_res["hits"][2]["hybrid_search_info"]["rank_fusion_score"].get<float>());
}
TEST_F(CollectionVectorTest, HybridSearchOnlyVectorMatches) {
nlohmann::json schema = R"({
"name": "coll1",